When you attach a picture of a design and ask the editor to make a section look like it, the reference now drives the visuals — background colour, palette, and signature effects like glows and light trails — instead of being toned down to fit the site's existing theme. Your approved wording, prices and links are still preserved exactly. **Where:** Site editor → click a section → attach an image in the chat → ask for it to be recreated from the reference. **How to test:** 1. Open a design, click a section (for example the hero), and attach a screenshot of a design you want it to look like. 2. Ask for the section to be recreated to match the attached image. 3. The result should adopt the reference's background and colour treatment, not keep the previous section's palette. 4. Effects that are obvious in the reference (glows, gradients, moving light) should be clearly visible, not faint. 5. Headlines, prices and button labels should be unchanged unless you asked for new copy. **Watch for:** A section that keeps its old colours despite the reference being clearly different; effects so subtle you have to look for them; or copy quietly rewritten to match the reference's own wording. If the reference contains photos or screenshots the editor cannot source, it should now tell you in its summary rather than quietly substituting a coloured panel.
Clients can now leave comments on a shared site preview, and their comment stays on the page once you've addressed it instead of disappearing. **Where:** Site editor → Share → copy the preview link → open it and click "Annotate", then click anywhere on the page. **How to test:** 1. Open a site with more than one design, click Share, and confirm "Allow annotations" is on (it's now on by default). 2. Copy the preview link and open it in a private window. 3. Click "Annotate", click any heading or button, type a comment, add your name and email, and submit. The pin should appear immediately with a number. 4. Switch to a different design in the preview and leave a second comment there. 5. Back in the editor, open Share → Feedback. Both comments should be listed, each against the design it was left on. 6. Mark one as Resolved, then reload the preview link. The pin should still be there, greyed out — not gone. **Watch for:** - A comment failing to save at all, or the pin not appearing after submit. - Two comments left on different designs being reported against the same design. - A resolved pin vanishing from the page instead of turning grey. - Pins landing in the wrong spot after you've edited the page — they should snap to the element that was commented on, or show a dashed outline if that element no longer exists. - Comment reviewers' email addresses should only ever be visible to you and your team, never to other people holding the preview link.
When you asked the AI to rebuild a section in the style of a reference image, it could keep one of the reference's own words in the design and show it next to your headline — producing a sentence you never wrote and could not edit. That leftover text is now tied to your own accent-line content instead, so the headline always reads exactly what your content says. **Where:** Site editor → select a section → paste or attach a design reference and ask the AI to match it. **How to test:** Pick a reference whose headline has a coloured last line, ask the AI to recreate the section in that style, then check the rendered headline. It should show only your headline (plus your own accent line if you have one) — no words carried over from the reference. **Watch for:** An accent line that disappears when you actually wanted it — if so, set it in the section's headline accent field and it will render.
Right after a new design was generated, the preview could show the new headline and text inside the *previous* design's layout for as long as it took to compile — which made it look like the AI had ignored your instructions. The preview now waits for the new design before updating the text. Prompt boxes also keep their button beside the input in custom hero designs, and motion instructions you write now outrank the built-in animation defaults. **Where:** Site editor → Design tab, after generating or regenerating a design. **How to test:** 1. Open a site with several designs and generate a new one from the Design tab. 2. Watch the preview from the moment generation finishes through the "Compiling…" state. 3. The preview should hold the previous design intact, then switch to the new design complete — never the new wording inside the old layout. 4. Generate a hero with a prompt/search box styled as a card. The button should sit inline beside the input, not wrap underneath it on the left. 5. Ask for continuous background motion (e.g. "flowing light trails across the hero"). The result should implement that motion, not just fade-in-on-scroll reveals. **Watch for:** A design that never leaves "Compiling…", text edits that appear not to apply until the preview refreshes, or a prompt box whose button overlaps the input on narrow screens.
When you attach a reference and ask the editor to recreate a section, it can now pull from images already uploaded to your site to fill photo areas the reference shows — screenshots, product shots, gallery tiles. Previously those areas came out as flat coloured boxes, which read as broken on a gallery or showcase. If nothing in your library fits, the editor now says so in its summary instead of quietly shipping a placeholder. **Where:** Site editor → click a section → attach a reference image → ask for it to be recreated. **How to test:** 1. Upload a few images to the site so the library isn't empty. 2. Attach a reference design that clearly contains photos, screenshots or a gallery. 3. Ask for the section to be recreated to match it. 4. Photo areas should contain real images from your library, not empty coloured panels. 5. If the reference needs imagery you don't have, the reply should name what's missing. **Watch for:** An image used somewhere it doesn't make sense (wrong subject for the slot), a gallery still rendering as blank boxes with no explanation, or the editor claiming it's finished when photo areas are clearly empty.
Client comments on a shared site are now a conversation: you reply, they get an email, and your answer sits on the pin they left it on. Clients you invite to the portal can comment too, and all feedback has one home in the editor. **Where:** Site → Feedback (new tab, badged with the open count). Also in Share → Feedback, and in the bell menu when a comment arrives. **How to test:** 1. Open a site, click Share, copy the preview link, and open it in a private window. 2. Click "Annotate", click a heading, and leave a comment. A first-time reviewer should see a one-off note explaining that links are paused while commenting, and Esc should get them out. 3. Back in the editor, go to Site → Feedback. The comment should be listed under the design and page it was left on, with an open count on the tab. 4. Type a reply and click "Reply & mark addressed". The reviewer should get an email; opening its link should jump to that comment. 5. In the private window, reload. The comment should show your reply, a badge on the comments button, and a "your comments have updates" prompt. 6. Add a follow-up as the reviewer. It should reopen the comment and email you. 7. Pick two open comments, click "Fix with AI", and check they land in the editor's pending changes for you to review — they must not apply on their own. 8. Invite someone to the portal as a reviewer, then open Review from the portal rail. Their name and email should already be filled in. **Watch for:** - Nine comments in one sitting should send you one email, not nine. - A dismissed comment should not email the client. - Comments should only be replyable by the person who raised them. - Reviewers' email addresses must stay visible to you and your team only, never to other people holding the link. - The comment list covers the whole site, but pins on the page should only ever be that page's.
The "Create a new design" screen now shows an editable plan beside the chat, so you can check what we understood — and fix it — before spending a generation. **Where:** Site editor → Designs rail → **+ New** **How to test:** 1. Open a site with a few designs and click **+ New**. 2. Describe a design, e.g. "hero like D4, footer like D5, keep everything else from D6, and make the hero title bigger." 3. In the plan on the right, check the section cards. Sections taken from an existing design show a real preview of that section with your current content; a section that has to be created shows a labelled placeholder instead. 4. Click **Change source** on a card and pick a different design — the preview re-renders. Designs that don't contain that section are listed but greyed out. 5. Click **Edit** on the design direction or any content value, change it, and Save. It gets an "Edited by you" marker. 6. Send another chat message and confirm your edited field is unchanged, while other rows update and show an "Updated" badge. 7. Generate. When it finishes, use **Refine with text** to carry on refining the new design instead of starting over. **Watch for:** Content rows should show the actual proposed copy, not just field names like "Title, Subtitle". The design direction should never be cut off mid-word. A section card whose source design lacks that section should say so rather than quietly using a different one.
Sites connected to their own authentication system now keep logged-in calls to action pointed at the owner’s configured dashboard. If that destination is missing, the original button is left unchanged instead of falling back to Mondovo’s dashboard. **Where:** Generated sites using endpoint, cookie, or browser-session authentication detection **How to test:** Configure customer-owned authentication with a dashboard URL, sign in, and confirm marked calls to action open that URL. Remove the dashboard URL and confirm those calls to action retain their original text and destination.
Prompt-driven hero sections now receive responsive, theme-aware form styling even when a generated design does not provide custom control classes. Text areas no longer show native resize chrome, and the action button remains visible and clearly separated from the prompt field. **Where:** Generated site heroes that ask visitors to enter a brief, search, or prompt before continuing **How to test:** Generate a site with a prompt-entry hero, then check desktop and mobile previews. The field should fill the available width, the button should have clear contrast and spacing, and submitting should open the real action destination rather than jump to another section.
Asking for more creative alternatives no longer jumps to a different section. Suggestion threads remember which section you were talking about, so follow-up pills and “more ideas” keep working on that section. **Where:** Site editor → Design Assistant chat (suggestion / clarification pills) **How to test:** Ask for alternatives on one section (e.g. objections), then ask for “something more creative” or click a few suggestion pills — edits should stay on that section, not another (like a steps/process block). **Watch for:** After a successful apply, the sticky subject clears; a new ask on a different selected section starts clean.
Videos built in Creative Studio sometimes had a line of supporting copy printed straight through the headline above it. We now catch that before the video is built and give the composition a chance to fix itself. **Where:** Creative Studio → Video **How to test:** Build a new video (or use Rebuild video on an existing one) and step through it with the scrubber. Each beat's headline and its supporting line should sit clear of one another. **Watch for:** Text that still touches or overlaps at any point in the timeline, or a beat whose supporting line has drifted unusually far from its headline. Existing videos are unchanged until you rebuild them.
An uploaded clip can now be dressed two ways and hold on to both. Captions and graphics over your footage is one; a presentation, where authored slides become the picture and you are composited back on top, is the other. Switch between them as often as you like — nothing is rebuilt, nothing is charged, and neither one is thrown away. **Where:** Creative Studio → Video → open a clip → the chooser at the top **How to test:** 1. Open a clip that already has captions on it. It should look and behave exactly as it did before — the chooser is there but nothing has changed underneath. 2. Build a presentation. The view moves to it and the export switches to it. 3. Switch back to "Captions & graphics". Your caption lines, your preset and your graphics are all still there, untouched. 4. Switch to "Presentation" again. Your sections and slides are still there too. Neither switch costs a credit or rebuilds anything. 5. Change a caption preset while both exist, then look at the presentation. It is unaffected. 6. Check the Export panel: when both treatments exist it names which one is going into the file. 7. Export, and confirm you get the treatment the chooser says is live. **Watch for:** a switch that appears to work but reverts a moment later; the Export panel naming one treatment while the file contains the other; captions or slides going missing after a switch; being charged for switching; a clip that has only ever had captions showing an export note it does not need.
Opening Visuals for a brand that had already made something dropped you straight onto the results of that old batch, with nowhere to type — now you land on the prompt, and the Ideate / Refine / Results labels are clickable so you can move between them. **Where:** Creative Studio → Visuals (and Brand Studio → Create) **How to test:** Open Visuals for a brand that has generated at least one visual before. You should see "What should we make?" with a text box, not the previous batch. Click "Results" to bring the last batch back, then "Ideate" to return to the prompt. If a generation is still running when you reload, it should still resume on Results. **Watch for:** Landing on Results with no text box; a running generation that no longer resumes after a page reload.
Uploads across the whole product — wizard logos and documents, chat attachments, plan images, and file fields on your published sites — now retry automatically instead of failing on a momentary storage blip, and tell you clearly when something genuinely didn't go through. Three of these paths were worse than a visible failure: they never checked whether the upload succeeded at all. A dropped logo or document was recorded as if it had worked, leaving a dead link in your site's content with nothing to indicate a problem. Those now report the failure properly. **Where:** Site wizard (Business basics, Content, Plan, Brief), chat attachments in the builder, image preference cards, and file-upload fields on published sites **How to test:** Upload a logo or document in the wizard and confirm it appears. Attach an image in builder chat. On a published site with a form that takes a file, submit an attachment and confirm it arrives in the Inbox. **Watch for:** A logo or uploaded document that shows as successful but renders broken later — that was the old failure and should no longer be possible. Also watch for uploads that now surface an error where they previously appeared to succeed; that is the fix working, but the underlying upload still needs to go through on a retry.
The "Generate 4 directions" button in Creative Studio's Logo Maker went back to looking idle and clickable the moment your request was accepted, even though the directions were still being drawn — so a second click could start (and charge for) another batch while the first one was still running. **Where:** Creative Studio → Logo → Direction brief **How to test:** Enter a direction brief and click Generate. While the direction cards read "Generating SVG…", the button should read "Generating 4 directions…" with a spinner and stay unclickable. Once every card finishes (or fails), the button returns to "Generate 4 directions" and you can run another batch. The refinement and Launch Kit starter-layout buttons behave the same way. **Watch for:** A button that stays stuck on "Generating…" after every card has finished, or one that becomes clickable while cards are still spinning.
A review of the new presentation mode found several things that would have gone wrong the first time anyone used it. All fixed before they reached you: portrait clips recorded on a phone now work, presentations export at the same resolution captions do, and the check that stops a slide covering the speaker no longer refuses videos it should have accepted. **Where:** Creative Studio → Video → open a clip → Presentation **How to test:** 1. Upload a clip recorded in portrait on a phone. Build a presentation and export it. It should render, and the file should be the same resolution as the clip you uploaded — not half of it. 2. Build a presentation whose slides use your brand colours rather than a dark background. It should export normally; it used to be refused. 3. If a slide does end up drawn over you, the export still completes and the render row carries a note naming the section, rather than failing. 4. Build a presentation and let the slide authoring fail (rare). You should not be charged, and you should be able to try again. 5. Change a caption preset and have it fail. Your previous captions should still be there and still exportable. 6. With both a presentation and captions built, look at the line under the mode cards — it should name whichever one actually exports, even when you are looking at the other. **Watch for:** a portrait export that comes out sideways or squashed; an export smaller than the clip you uploaded; a presentation refused with a message about a section where nothing is drawn; being charged for a build that failed; a note claiming the version you are looking at exports when the badge says otherwise.
A section could go blank in the editor preview when its call-to-action was stored as a group of fields (heading, sub-heading, button label, button link) while the design rendered it as a single line of text — now the button label is shown instead, and new designs are checked for the mismatch before they are saved. **Where:** Site editor preview — any section with a call-to-action or an address, e.g. `/client/{clientId}/site/{siteId}` **How to test:** Open a site whose content has a call-to-action group on a section (heading + button label + button link). Load the editor preview and confirm the section renders with the button label as the link text rather than collapsing to an empty block. **Watch for:** A section that still renders empty, or a link that shows the call-to-action heading instead of the button label.
Suggested graphics can now reach for a drawn symbol from a library of 236 — a padlock, a stopwatch, a wrench, a delivery van — or put a real photograph on screen beside you. It looks in your own pictures first, then free stock, and only makes a new one if neither has what you described, so most of it costs nothing extra. **Where:** Creative Studio → Video → open a clip → "What should the graphics say or show?" → Suggest graphics **How to test:** 1. Open Creative Studio → Video and upload a clip where you name concrete things — a dish, a tool, a place, a delivery. 2. Click **Suggest graphics**. 3. Rows should now include symbols ("Symbol: lock at 0:09") and photos ("Photo: a rustic sourdough loaf… (a stock photo) at 0:17"). The bracket tells you where the picture came from — your own pictures, a stock photo, or made for this. 4. Play the preview. A symbol assembles piece by piece; a photo wipes up into a bordered card and drifts very slightly while it is on screen. 5. Check the credit line on the button, then check what you were actually charged. A plan of symbols and drawn accents is one credit. Only a made-for-this illustration or a made-for-this photo adds one. 6. Export. Both are baked into the file exactly as previewed. **Watch for:** a photo card covering your face; a symbol that has nothing to do with what is being said; a card that renders as an empty bordered box; a photo that appears in the list but not in the player; being charged more than one credit when every row says "from your pictures" or "a stock photo"; the same stock photo appearing on two different cards in one clip.
Comparing narration voices is now one tap — click any voice and it plays instantly. **Where:** Creative Studio → a video project → Voiceover panel (both the Recommended cards and the All voices drawer). **How to test:** Open the Voiceover panel, tap a Recommended voice or open All voices and tap one — a short sample plays immediately. The "Preview voice" button still reads your exact accent, pace and direction. **Watch for:** The tap sample is a fixed neutral line; it will not reflect a non-default accent/pace/direction (the Preview voice button covers that). Playback should start on the tap and remain a scrubbable player if a browser blocks autoplay.
When an AI edit quietly switches a section to read a different page's content, the editor now says so instead of reporting a clean "Done" over a section that has gone blank. **Where:** Site editor → AI chat panel, after any refinement that changes code. **How to test:** Ask the assistant for a change to a section that is backed by a list (portfolio, menu, services). If the edit repoints that section at another page's content, the reply now carries a warning naming the old and new content, which fields no longer match, and an offer to point it back. **Watch for:** The warning should only appear when a section genuinely lost data it used to show. An edit that deliberately switches a section to different content and rewrites it to match should complete with no warning at all.
Upload a video of yourself talking and we will build the slides to go with it. We read what you actually said, break it into sections, write each slide from your own words, and put you back on top — beside the slide while you are making the point, or in a corner circle while a chart is doing the work. **Where:** Creative Studio → Video → open a clip → write out the captions → "Turn this into a presentation" **How to test:** 1. Upload a clip of someone talking for 30 seconds or more — one that makes a few distinct points, ideally with a number in it. 2. Write out the captions first. A presentation is built from the script, so the button stays disabled until the words exist. 3. Optionally say where you are in the frame (left / middle / right) and add a line about what it should focus on. Leave the position on "Not sure" and we work it out from the video. 4. Click through to build. It takes a couple of minutes — it plans the sections and then designs the slides. 5. Review the sections: each shows its timecode, its title, its bullets, where you sit for it, and why it is a section. 6. Edit a title or a bullet, then rebuild. Your edits are kept — it does not re-plan from scratch. 7. Export. You should be composited cleanly onto the slides, moving between positions on the section changes. **Watch for:** a slide drawn where you are standing (we now refuse the render rather than ship it, so you would see an error naming the section instead); the bubble showing an empty room instead of your face; you disappearing between sections; slides that repeat your words verbatim instead of summarising; the credit charge exceeding three. Building a presentation replaces whatever captions or graphics were layered on the clip — going back the other way now asks you to confirm first instead of silently discarding the slides.
When you select a specific element (like a single link or button) and type a deictic style/color change such as "make the background black and the text white", the change now applies to just that element instead of restyling its whole section. **Where:** Site builder → select an element in the preview → type a color/style change in the chat. **How to test:** Open a site, click a link inside a section (e.g. a "Shop Collection" link in the Hero), then type "make background colour black and text white". Only the selected link should change; the rest of the section stays as-is. **Watch for:** If your prompt names a different section by name, that named section still wins — an old/stale selection will not hijack the edit.
Extended the video overlap check so an opaque card, panel, or image can no longer be placed on top of a headline — even when the two live in different parts of the scene. **Where:** Video generation (Creative / Motion Studio) **How to test:** Generate a new video whose scene shows a headline above a prompt card, mockup, or image panel — the graphic now sits below the text instead of covering its last line. **Watch for:** A freshly generated scene where a solid card or image still overlaps live text.
Uploading your own footage no longer loses the whole file when storage has a momentary blip — the upload is retried automatically, and if it still can't go through you get a message that says what happened. **Where:** Creative → Video → Upload & edit (also the video pickers in the site editor: Replace media, and the stock-video "Upload" tab) **How to test:** Upload an MP4/WebM/MOV of 20 MB or more and let it run to completion. It should land and open normally. Re-upload a few times — an occasional storage hiccup now shows "Upload hiccuped — trying again…" and then finishes, instead of stopping. **Watch for:** A failed upload that reports a bare code like "Upload failed (502)" — that message should no longer appear. Also watch for an upload that appears to hang: the button text should always name the current step (checking / uploading / retrying / opening).
Multi-cut videos now build their scenes together instead of one after another, every change you ask for is kept with what it actually did to the video, and the recorded voiceover lines play properly instead of showing an empty player. **Where:** Creative Studio → any motion video. Building happens on Generate; the history sits under "Change something"; the players are in the Voiceover panel. **How to test:** 1. Generate a video with 4–6 cuts and watch the progress card — it should count "N of M scenes done" and finish in roughly a third of the time it used to. 2. Open "Change something" and ask for something. Under the box you should see your request with a line beneath it saying what happened — which cuts changed, or that it was left alone. 3. Ask for something deliberately impossible ("replace the video with a photo of my dog"). It should be recorded as declined rather than vanishing. 4. Open the Voiceover panel: every recorded line should show its real length in the player, not 0:00 / 0:00, and play on click. 5. Preview a voice, then collapse and reopen the panel — the preview should still be there. **Watch for:** a progress count that goes backwards or sticks; a video whose scenes no longer share one look (colours or fonts drifting between cuts); a change history that disagrees with what the video actually looks like; a build that fails partway and still charges credits.
Sections created from the Add section panel now receive your site's actual colour tokens, so they match the rest of the page instead of arriving in a generic palette. **Where:** Site editor → left rail → Add section (also applies to iterating on a saved block, and to the layout regeneration option in the section explorer) **How to test:** Open a site whose design uses a distinctive palette, add a new section from a short prompt, and check the generated section's backgrounds, buttons and accents against the surrounding sections. Then change a theme colour and confirm the new section moves with it rather than staying on its original colour. **Watch for:** A newly added section that arrives in blues/greys unrelated to the site, or one whose colours stop tracking the theme after you change a colour — both mean it fell back to hardcoded values instead of the site's tokens.
In voice mode, the assistant's spoken reply and your own message used the same grey bubble colour, so on a phone the two could be hard to tell apart. **Where:** Site editor → Assistant panel, voice mode. **How to test:** Start a voice conversation with the assistant and ask for a change. The assistant's reply now appears in a distinct tinted bubble instead of the same grey used for your own messages. **Watch for:** The user's own bubble should stay grey — only the assistant's reply should change colour.
Ad creation now works the way people actually brief it. Upload a product photo right inside the create flow and the AI reproduces your actual product — shape, materials, colors — in whatever scene you describe, including quick "reshoot" chips (new angle, in use, on location, studio white). Every reference image you attach carries a role: **Product** (reproduce it faithfully), **Style** ("make an ad like this — but for my product": borrows the look, never the content), or **Setting** (shoot it in this place). And on any finished creative, **More like this** generates a fresh take on the same idea — different angle, light, or crop — for one credit. **Where:** Creative Studio → open a business → Create → Configure step (upload + roles + reshoot chips); results panel (More like this) **How to test:** 1. In Configure, click "Upload a product shot", pick a distinctive product photo, and confirm it appears selected with the Product role. Generate — the product in the ad should match yours faithfully, not a lookalike. 2. Attach a second image and set its role to Style; the result should borrow that image's mood and lighting without copying its subject or any text. 3. With a product reference selected, click a reshoot chip — it appends to your visual direction rather than replacing it. 4. On a finished creative, click "More like this" — a new "Take N" card should appear with the same concept executed differently, and one credit charged. **Watch for:** an upload that lands in the library but never appears in the picker; a style reference whose subject or text leaks into the output; "More like this" returning a near-duplicate; roles not surviving a retry of the same job.
Three fixes to the results step for creatives, found while testing the new carousel flow. **Where:** Creative Studio → Visuals → after a creative finishes generating **How to test:** 1. Create a carousel and let it finish. The deck appears with its slide strip — even if you cleared the headline, offer and call-to-action fields at the review step. Previously a carousel with no copy in those fields sat behind a "preparing" spinner forever, with no way to view, download or delete the creative you had just paid for. 2. On any finished creative, open Edit, change the copy, and apply it. Now use Copy URL, Approve or Save to Website. These act on the image you are looking at. Previously they acted on the version from before your edit — a working link to the wrong picture. 3. Reload the results page a few times, then open your brand library. Each slide appears once. Previously every reload saved another copy of every slide, and deleting the creative left the extras behind. **Watch for:** a carousel that finishes but shows no slide strip; a copied image URL that does not match the creative on screen; repeated copies of the same slide in the brand library.
You can now make a full carousel — a cover image plus as many typographic slides as your argument needs — and it costs the same as a single post, because only the cover is generated. **Where:** Creative Studio → Create → pick **Carousel**. Or open a planned carousel straight from the Calendar: click a post whose format is Carousel and choose *Create visual* — it now arrives as a carousel instead of a single image. **How to test:** 1. Go to Creative Studio → Create and choose **Carousel**. 2. Pick **3–6 slides** and a shape (square or portrait), then write your visual direction and continue to Review. 3. Confirm. One image is generated; the remaining slides are built instantly from your brand colours and typeface — a title and bullet points on each, numbered in the corner, and a closing slide with your call to action. 4. On the result, use the slide strip to read each slide full size, the arrow keys to swipe through them, **Download all** to save the set in order, and **Edit** to change any slide's wording and rebuild. 5. Delete removes the whole set, not just the cover. **Watch for:** slides where the text is cramped or spills past the edges; a slide background that is too pale for the white text on it; slides arriving out of order, or the last slide missing its button; the cover image trying to draw words instead of leaving room for them; and Download all stopping partway through the set.
Sections that pull from a data source now appear in Settings, and you can hand-pick which items they show and in what order — no more asking the assistant and hoping. **Where:** Site editor → Settings → Data Sources → "Sections using live data" → Choose items **How to test:** Open a site with a section wired to live data. Confirm it is listed with its source name, click "Choose items", pick a few, reorder them with the arrows, and save. The preview and the published site should show exactly those items, in that order. Clearing the selection goes back to showing whatever the source returns. **Watch for:** A section listed as "needs attention" means its source is failing or was deleted. If the picker shows no items, the source returned nothing — check the connection itself. Sections filled by live data also now carry a "Live data" badge in the sections list.
When you connect Mondovo to an AI assistant, it can now actually look at your designs instead of reading a list of labels — and you can point it at one design and ask for another in the same style. **Where:** Any AI assistant connected to your Mondovo account (Settings → Connections), then ask it about a site. **How to test:** 1. Connect Mondovo to your assistant and ask it to list a site's designs. It should describe what each one actually looks like — the thumbnails now come through as images, each captioned with its design number. 2. Ask it to check a design for problems. It sees the rendered page alongside the error and contrast report, so it can spot a broken hero or invisible header on its own. 3. Ask for "another version like design 2, but with a shorter pricing section". It generates a sibling of that design — same colours, type and personality, different layout — instead of an unrelated new direction. 4. Ask plainly for "another option" with no design named, and you still get a deliberately different one. **Watch for:** Designs generated before today were stored as narrow, low-detail screenshots, so their thumbnails may look soft until they are re-captured; new designs are stored at full width. A design created seconds ago has no screenshot yet, and the assistant will say so rather than guessing. Site code is still never shared — the assistant works from pictures and generates on our servers.
What you say to the Creative Studio assistant is now saved as a conversation, the same way the site assistant already worked. Before this, a voice call in the studio ended and the conversation went with it. **Where:** Creative Studio → any brand → "Talk to the studio". **How to test:** 1. Start a voice session in a brand's workspace and talk through a couple of requests. 2. End the call. 3. The turns are kept against that brand rather than discarded — a second session with the same brand continues the same thread. **Watch for:** - Ending a call twice (closing the tab mid-session can send the report twice) should not duplicate the conversation. - Someone who is not a member of the brand's organization should never have a conversation created for them.
When an AI assistant checks a design for problems, it now only sees errors that actually belong to that design — and is told when an error was recorded before the design was last edited. **Where:** Any AI assistant connected to your Mondovo account, asking about a specific design. **How to test:** 1. Ask a connected assistant to check a design that has had errors in the past. 2. Errors recorded while a different design was on screen no longer appear, and the reply says how many were excluded. 3. Errors older than your last edit to that design are marked as possibly already fixed, so the assistant verifies before changing anything. **Watch for:** Errors from a published page carry no design marker, so they are listed as belonging to the site rather than to one design — that is expected. Contrast warnings are advisory and are not automatically rewritten; a low ratio on large or decorative text is often fine.
"Download all" on a carousel take now delivers every slide in a single .zip instead of only the first image. **Where:** Brand Studio → Create → a carousel take → Download all **How to test:** Generate a carousel (3-6 slides), then click "Download all". You get one .zip containing every slide, named in deck order. **Watch for:** Previously browsers blocked the burst of downloads and only the hero slide saved; a slide that failed to fetch is now skipped and reported rather than aborting the rest.
You can now drop a video onto the upload screen instead of hunting through a file dialog, and attach your own photos to the creative chat so every idea it proposes is built around them. **Where:** Creative Studio → Video → Upload & edit (drop a clip anywhere on the upload panel), Creative Studio → Video → Create a video (drop images and clips onto the Assets panel), and Creative Studio → Visuals (the "What should we make?" chat). **How to test:** 1. Open Creative Studio → Video → Upload & edit and drag an MP4, WebM or MOV from your desktop onto the upload box. It should highlight as you drag over it and start uploading the moment you let go. 2. Drag a non-video file onto the same box — it should refuse it with a plain message rather than opening the file in the browser. 3. Go to Creative Studio → Visuals. Drag a photo from your desktop anywhere onto the "What should we make?" panel. It uploads, joins your brand's images, and shows as an attachment above the message box. 4. Click "Attach images" to pick pictures you already have, or to upload one. Up to four at a time. 5. Press send without typing anything. The idea that comes back should be built on the pictures you attached — check the reference thumbnails on the proposal card. 6. Remove an attachment with the small × and send again; the next idea is free to go elsewhere. **Watch for:** a drop that opens the file in the browser tab instead of uploading; an attached photo that never shows up on the proposal card; the send button staying greyed out when images are attached but no text is typed; attachments surviving a page refresh (they are not meant to — the pictures a proposal was built on carry forward in the proposal itself).
After the AI edits your design, the chat now answers in one clear sentence instead of a dense paragraph — with the details one click away, and a button that jumps the preview straight to the section that changed. **Where:** Site editor → Assistant panel (right side), after any design change **How to test:** 1. Open a site in the editor and ask the assistant for a change ("make the objection cards two columns"). 2. When it finishes, the reply should be a single short sentence — no design theory, no "consider also…" tail. 3. Click "Show details" to see the specifics as a bulleted list. 4. Click "Show me what changed" — the preview scrolls to the edited section and pulses it. Hovering the button highlights it without scrolling. 5. If you had an element selected but named a different section in your message, a small line says which section was edited, with a one-click "Use my selection instead". **Watch for:** Replies that still run long or repeat the follow-up suggestions that already appear as clickable chips below. Also watch for "Show me what changed" pulsing the wrong section, the whole page, or nothing at all — and for long conversational answers (questions you ask the assistant, not edits) getting cut short, which should not happen.
The results screen now leads with the picture instead of a wall of buttons. Every finished take is a clean tile with four things you can do: **View**, **Edit**, **More like this**, and **Delete**. View opens a full-screen viewer where you can flip to original size, page between takes with the arrow keys, and download the file. Everything that changes the creative — copy, layout, logo, regenerate, favorite, approve, save to a website, export link — now lives inside Edit, where you can see what you are changing. Takes save themselves to your brand library the moment they finish, so there is no "Save to Brand" step any more; if you do not want one, delete it. And each take in a batch now gets its own composition, so two different photos no longer come back with the headline in exactly the same place. **Where:** Creative Studio → open a business → Create → generate a batch → results **How to test:** 1. Generate two or more variants. When they finish, each card should show only View, Edit, More like this and Delete — no Save to Brand. 2. Compare the two takes: the copy should be laid out differently (for example a bottom band on one, a top band on the other), not identically. 3. Click View. The creative should fill the screen fitted; click it (or "Original size") to see it at full resolution and scroll around; use ← and → to move between takes; press Esc to close; click Download and confirm a file saves. 4. Open the brand library — both takes should already be there without you saving them. 5. Click Delete on one take, confirm, and check it disappears from the results and from the library. 6. Click Edit and confirm the copy fields, layout presets, logo toggle, regenerate, approve, Save to Website and export link are all there. **Watch for:** a take appearing twice in the library; Delete removing the card but leaving the library copy; the viewer trapping keyboard focus or failing to restore it on close; a layout you chose in Edit reverting to the automatic one.
On sites that happened to have a content collection named "sections", asking the assistant to add a new page section could be misread as a request to create that collection — so the chat said "created", but the section you actually asked for never got built. **Where:** Site editor → chat → "Add a new section..." **How to test:** - On a site with an existing "sections" content collection, ask the assistant to add a new section (e.g. an animated block) - Confirm the section is actually designed and added to the page, and the chat doesn't just report a collection being created **Watch for:** The chat replying that a collection was "created" when you asked for a page section — that's the exact failure this closes.
Some sites had accumulated duplicate copies of the same page file, which made the AI editor wrongly report that a change "would have removed another section" and refuse every refinement. **Where:** Site editor → AI refinement (any "add/edit a section" request) **How to test:** Open an affected site and ask the assistant to add or edit a section; the change now applies instead of being refused. **Watch for:** The duplicate copies are cleaned up automatically the next time a refinement is saved.
When the assistant added a brand-new section and you then asked it to change that same section, the change was saved but the preview kept showing the original version — so it looked like nothing happened, no matter how many times you tried. **Where:** Site editor → Assistant panel, after asking for a new section (e.g. "convert this into a left/right section") and then refining it ("make it less text heavy", "try again") **How to test:** 1. Open a design and ask the assistant to add a new section. 2. Once it appears, ask for a change to that same section — without reloading the page. 3. The preview should update immediately. Repeat two or three times; every round should be visible. **Watch for:** A follow-up that reports "Applied" while the section on screen stays identical. If that happens, reload the page and note whether the reloaded version shows the change — that would mean the edit reached the database but not the preview.
On a phone, the design editor was leaving a tall blank band under the site preview whenever the "Undo last change" pill was on screen — which is most of the session, since that pill stays until the change is undone. The preview now reaches almost to the bottom edge again. **Where:** Site editor on a phone — Design view (the screen with the D1/D2/D3 design tabs). **How to test:** Open a site's design view on a phone, make any AI change so the "Undo last change" pill appears, then close the chat. The pill now sits on the same line as the chat button instead of stacked above it, and the site preview fills the space that used to be blank white. **Watch for:** The Undo pill or the status pill drifting on top of the site's own buttons, or the pill colliding with the round chat button in the corner.
Swapping a section's layout often showed only one option, previewed it in a tiny cropped strip, and started generating the moment you clicked. Browsing is now free, previews show the whole section, and layouts that would render empty can no longer be applied. **Where:** Site editor → select a section → Quick Actions → Show Alternative Layouts **How to test:** 1. Open the picker on any section. You should see a fuller set of designs than before, including ones adapted from related section types. 2. Click a design — it now just selects. Press "Adapt this design" / "Build this layout" when you actually want it generated. 3. When the result appears, the card shows the whole section rather than the top sliver. Use "Full size" to inspect it properly before deciding. 4. Apply it. The section content should carry over intact. **Watch for:** A card marked "Would render empty" with its button disabled — that means the generated layout reads the wrong content field, and applying it would have given you a blank section. Generate that option again rather than forcing it.
If the assistant created a new section whose name matched something already in your design (say, an old helper or an earlier version of that section), the new section used to be silently dropped and never appear. It's now given a safe new name up front so it saves and shows on the page as expected. **Where:** Site editor → chat, any edit that adds a new section **How to test:** - Ask the assistant to add a section, then ask it to add another section that would plausibly reuse a name already used in the design - Confirm the new section actually appears on the page rather than silently disappearing **Watch for:** A requested section not showing up in the preview after the assistant says it was added.
Asking the assistant to make a selected heading bigger or smaller sometimes only changed its phone size, so nothing moved on the desktop preview you were looking at — even though the assistant reported success and offered an Undo. The resize now carries across every screen size, keeping the heading's original size ramp intact. **Where:** Site editor → select a heading in Edit mode → ask the chat to make it larger **How to test:** Open a site, click a section heading so the chat shows "What would you like to change about H2: …", send "Make this heading a bit larger", and watch the preview at the default desktop width. The heading should visibly grow immediately. **Watch for:** A heading that grows on mobile but not on desktop (or vice versa), or one that jumps by more steps than you asked for. Switching the device preview to Mobile and back should show a consistent, proportional size increase at both widths.
Asking the assistant to make a specific selected element bigger or smaller could come back reporting the change was made while the element stayed exactly the same size — or, more recently, come back saying it could not make the change at all. Both symptoms had the same cause, and edits like this now land. **Where:** Site editor - click an element in the preview, then ask for the change in the chat panel. **How to test:** Open a site in the editor, click the main hero headline to select it, and send "Make this bigger." The headline should visibly grow, only that headline should change, and the change should survive a reload. **Watch for:** The assistant claiming a size change that did not visibly happen; other headings, buttons or sections in the same area changing size when only one element was selected; the assistant asking you to rephrase a request it used to handle.
When you asked the AI to add a new section and then refine it, the two steps could each build their own copy of the section — one visible on the page and one saved as a hidden duplicate. Later edits sometimes landed in the hidden copy, so the chat said "saved" but the page never changed, even after a refresh. The AI now coordinates the two steps so only one copy exists, and a safety net removes any hidden duplicate before saving. **Where:** Site editor → Design chat (any request that adds a section, then edits it) **How to test:** Ask the chat to add a new section (e.g. "add an FAQ section"), then immediately ask to restyle that same section ("make it two columns with icons"). The restyle should be visible on the canvas right away and survive a page refresh. **Watch for:** A refinement that replies "I saved that change, but the live preview couldn't update" and shows nothing after refresh — that was the exact symptom of this bug.
A section wired to live data could fail to render if any incoming row left a field blank — the safety check meant to prevent a display glitch was itself crashing on empty values. **Where:** Site editor → any section bound to live data (for example a showcase pulling from the template gallery) **How to test:** Open a design whose section pulls items from live data, and confirm the preview renders with all items — including ones where a field like category is empty. Existing designs heal automatically the next time their code is fixed up. **Watch for:** A section that renders blank filter tabs or empty labels — that means the binding is reading a field that is unset on most rows; point it at a field that is always present.
The review step now shows every concept as a real mock — your headline, offer and button laid out exactly as they will be composed, with a grey placeholder where the photo will go — so you judge a design instead of imagining one. **Where:** Brand Studio → Create → pick an ad, banner, social post or carousel → Review creative **How to test:** Start a creative with a couple of concepts and continue to Review. Each concept card now carries a small mock: the copy set in the real layout over a grey photo placeholder. Open "Edit copy & prompt" and change the headline — the mock re-draws about half a second after you stop typing. Click the grey placeholder on any card to edit what the image should show, hit Save, then generate: that concept is generated from the direction you just wrote. Carousels preview their cover slide. **Watch for:** A mock that never appears or stays behind a shimmer; copy that renders differently in the finished ad than it did in the mock; the image direction you saved not being the one used for that concept; the mock reading as a finished design rather than an obvious placeholder.
Reviewing a carousel now shows the whole deck as it will look — every slide, in order, with the copy set exactly as the composer will set it — and you can edit any slide's copy right there. **Where:** Brand Studio → Create → pick a carousel → Review your creative **How to test:** Start a carousel, continue to the review step, and check that you get one small preview per slide instead of just the cover. Click a slide to open its fields: the cover has a headline, the middle slides have a title and their bullet points (add or remove points within the allowed range), and the last slide has a closing line and a call to action. Type a change and watch that slide's preview redraw a moment later. Generate, and the finished slides should match what you saw. **Watch for:** The middle slides carry no photograph, so their previews are the finished slides — the colours and gradient should match the composed result exactly. The cover and the last slide still show the grey placeholder where the one generated image lands; clicking that placeholder is still how you change what the image shows. Long copy is not blocked while typing — the counter warns you, and anything over the limit is trimmed when you generate.
Choosing a new layout from "Show Alternative Layouts" could leave the design unable to render at all — and because the swap was never recorded in version history or the chat, there was no way to undo it. **Where:** Site editor → select a section → Quick Actions → Show Alternative Layouts → Use this layout **How to test:** 1. Select a section and open Show Alternative Layouts. 2. Pick a layout and apply it. The section should change and the rest of the page should stay intact. 3. Check the chat — you should now see an "Applied a new layout to …" entry with an Undo button, and a matching entry in Version History. 4. Click Undo; the previous layout should come back. **Watch for:** A layout that cannot be applied now tells you why and changes nothing, instead of applying. If you see "Applying it would have broken this design", that is the guard working as intended — the design is untouched.
You can now work in the Creative Studio by voice — move between tools, ask what you have made, plan a post and have it generated, all without touching the keyboard. **Where:** Creative Studio → open any brand's workspace → "Talk to the studio" in the bottom-right corner. **How to test:** 1. Open Creative Studio and pick a brand. 2. Click "Talk to the studio" and allow the microphone. 3. Try "open the calendar", "what did I make last week", "switch to the playful voice". 4. Ask for something to be made — "an ad for the summer offer". You will hear the idea and what it costs, and a card appears with the same brief and price. 5. Say yes, or press "Make it" on the card. You will be told when the image lands, even if you have moved on to talking about something else. **Watch for:** - Nothing should ever be generated before you have heard the price and agreed — if a generation starts without a quote, that is a bug worth reporting. - Pressing Cancel on the card should mean nothing is spent, and the assistant should acknowledge it. - On a plan that does not include voice, the button should explain that rather than failing silently. - Team members with view-only access can ask questions and move around, but cannot make anything.
When you asked the assistant to change a headline or other wording, the change was saved but the preview kept showing the old text until you reloaded the page — now it updates on the spot. **Where:** Site editor → Assistant chat (any design) **How to test:** Open a design, click a heading in the preview, and ask the assistant to change its wording (typed or via a suggestion pill). The preview should show the new wording as soon as the assistant says it's done — no reload. **Watch for:** The rest of the section (list items, subtitles, images) must stay intact, and Undo should still revert the wording change.
When the Creative Studio chat suggests ideas that use your library images, it now assigns each one a job — your actual product to reproduce, a style reference whose look it wants to borrow, or your own place as the setting — and shows a small badge on the idea card so nothing surprises you at the review step. A style pick always comes with a one-line reason for what look it's borrowing. **Where:** Creative Studio → open a business → Create (chat suggestions and chat replies) **How to test:** Open a workspace for a business with a few library images. Idea cards that use an image whose subject isn't the ad's subject should badge it "Style reference" and the idea's rationale should say what look is borrowed. Picking the idea should land in Review with the same roles preselected on each reference. **Watch for:** a style-badged image whose subject or text shows up in the generated ad; roles shuffling when the chat refines an idea; more than one style reference on a single idea.
Opening Creative Studio or Motion sometimes showed "We could not come up with ideas just now — try again" even though a full set of ideas had been written — the answer just came back slightly misformatted and was thrown away. Those answers are now read properly, and if one is genuinely unusable we quietly ask a second time instead of showing an error. **Where:** Creative Studio → any brand workspace (the ideas shown when the panel opens); same panel in Motion. **How to test:** Open the creative or motion workspace for a brand several times in a row. Every open should land on a set of idea cards. **Watch for:** An idea card with blank or obviously wrong copy — that would mean a malformed answer was patched up when it should have been re-asked.
Sections with motion — an animated chat demo, counters that tick up on scroll — now come with their own controls, so you can adjust timing and behaviour directly instead of describing the change and regenerating. **Where:** Site editor → click an animated section → Advanced → Motion **How to test:** Create a section that moves — "an animated chat mockup where messages appear one at a time" or "stats that count up when scrolled into view". Select it in the preview and open Advanced. Under Motion you'll find controls the section defined for itself, named for what they do (message interval, entry style, loop). Drag a slider and the change saves to that section. Switch to another design variation and set a different value — each variation keeps its own. **Watch for:** Controls only appear on sections that actually animate; a static section shows the usual options and nothing extra. Sections created before this update won't have them until they're regenerated. A slider saves when you let go, not while you're dragging.
Your growth scan's content ideas are now a backlog you can schedule from — one click moves an idea onto your content calendar with its wording already filled in. **Where:** Site editor → Marketing → Ideas & Backlog → Backlog **How to test:** 1. Run a growth scan on a site so it produces content ideas. 2. Open Marketing → Ideas & Backlog → Backlog. It is called a backlog now, not a calendar — there is only one calendar in the product. 3. Expand a social post, video, infographic or promotion idea and click **Schedule this**. 4. Open the content calendar (the link at the top of the backlog, or Creative Studio → Calendar). The idea is there as a real post: its title became the hook, its copy starter became the caption draft, and its suggested date became the day. 5. Go back to the backlog — that idea now says "On the calendar" instead of offering to schedule again. 6. Edit the caption or drag the post to another day; the backlog is not affected. The post is the plan of record from here. **Watch for:** blog posts, emails and case studies deliberately do not offer scheduling — the content calendar is for social posts, and there is nowhere to publish the others yet. A website with no brand yet will tell you to open it in Creative Studio once first.
Creatives can now be made in a tall Pinterest pin (1080×1620) and a wide link card (1200×628), alongside the square, portrait, story and landscape formats — with headline, offer and CTA sized and positioned for each one. **Where:** Brand Studio → Create → Configure → "Channel format" (and the same picker on a finished result, where you can re-compose copy onto another format). **How to test:** Start a creative, pick "Pinterest pin (2:3)" or "Link card (1.91:1)", generate, then add a headline, offer and CTA. The finished image should come out at exactly 1080×1620 or 1200×628, with all copy sitting inside the frame and nothing clipped at the edges. **Watch for:** Copy that looks shrunken on the link card, a link card that arrives square or heavily cropped, or a subject that ends up hidden behind the copy band rather than above it.
Voiced videos no longer carry half a second of dead air in every cut, you can set the pace of the read, and a change you type can be aimed at the whole video instead of only the cut on screen. **Where:** Creative Studio → any motion video → the Voiceover panel and "Change something", both under the preview. **How to test:** 1. Open a video with a voiceover and look at the recorded lines — each clip now holds its line and almost nothing else. 2. In the Voiceover panel, pick a Pace (Relaxed / Natural / Brisk), re-record, then regenerate the video. Brisk should come back noticeably shorter for the same script; Relaxed longer. 3. In "Change something" on a multi-cut video, choose "Whole video" and ask for something visual — e.g. "switch to a dark background". Every cut should change, with the progress line naming the cut it is on. 4. Switch to "Just cut N" and confirm the next change lands only there. 5. Check the credit line above the button: it reads the real total for the whole-video option, not one credit. **Watch for:** a line whose audio is clipped at either end; a voice that ignores the pace you picked; a whole-video change that reports success but leaves some cuts untouched without saying so; the credit charged not matching the number of cuts that actually changed.
Your design conversation is now recorded as it happens, so it's there when you come back — including on a different device. Before this, the conversation only lived in the browser tab you were using. If you spoke a few changes on your phone and then locked it, or switched apps, the transcript could disappear entirely — even though the design changes themselves had saved correctly. Opening the same site on a second device could also roll the conversation back to an older state. **Where:** The chat panel in the site editor (Design view), on any device. **How to test:** 1. Open a site's design view and make a change by voice or by typing. 2. As soon as the change is confirmed, lock your phone or switch to another app (or just close the tab). 3. Re-open the same site — on the same device or a different one. 4. The request and the result should both still be in the chat, in order. 5. Make a change on a phone and a different change on a laptop, then reload both. Both should appear; neither should erase the other. **Watch for:** A request or result showing up twice in the chat; a turn filed under the wrong design (D1 vs D4) or the wrong page; the conversation appearing but out of order.
Emptying a carousel slide used to quietly delete it and shuffle the remaining slides into new roles, so the deck that got built was not the one on screen — now you are told which slide needs copy, and the editor always shows the deck that was actually rendered. **Where:** Brand Studio → Create → carousel → review step, and the slide editor on a finished carousel. **How to test:** Plan a carousel, clear the copy on any slide, then try to generate (or hit "Rebuild the slides"). You should get a message naming that slide instead of a silently different carousel. Put the copy back and rebuild — the slides on screen should match the built deck exactly, including any trimming of over-long lines. **Watch for:** A deck of fewer than three slides is now refused everywhere rather than accepted in one place and clamped in another; a plan that comes back short falls back to a single cover image instead of an unbuildable deck.
Suggested graphics used to be six kinds of text on screen. Now they can also be drawings — a reaction to how a moment feels, an arrow to what matters, ticks as things are listed, a burst on a punchline — and when nothing in that set fits, we draw something for the moment itself. **Where:** Creative Studio → Video → open a clip → "What should the graphics say or show?" → Suggest graphics **How to test:** 1. Open Creative Studio → Video and upload a short clip of someone talking. Captions are optional — this works on a silent clip too. 2. In "What should the graphics say or show?", ask for something wordless: *"illustrative graphics for the emotions in this, not text"*. 3. Click **Suggest graphics**. The list should come back with drawings in it, not only text panels — each row naming what it draws ("Expression: surprise at 0:06", "Illustration: a padlock clicking open at 0:17"). 4. Play the preview. Each drawing should assemble itself — strokes drawing on, parts popping in — land beside the speaker rather than over their face, and clear the caption band. 5. Switch one off and back on. The preview rebuilds both times, and a drawing that was made for this clip is still there when you switch it back on. 6. Export. The drawings are baked into the file exactly as previewed. **Watch for:** a drawing sitting on top of the speaker's face or inside the caption band; a graphic that appears fully drawn, then rewinds and draws again; a drawing still assembling itself as it fades out; a row that says only "Illustration" with no description of what it draws; an illustration you can see in the list but not in the player; the credit charge exceeding two.
Refreshing the page no longer throws away the conversation you were having with the Creative Studio or the Motion Studio — the thread comes back, and so does the idea it was working on. **Where:** Creative Studio → any brand → the chat panel in Visuals, and the "What should we make?" panel in Video. **How to test:** 1. Describe something in the chat and let it answer with an idea. 2. Refresh the page. 3. The conversation is there, and the proposed idea is still the one on the table — "Adjust details" opens it with the same fields. **Watch for:** - The Visuals chat should only show Visuals turns, and the Video chat only video turns, even though both belong to the same brand. - A conversation you have already started should never be replaced by an older one loading in behind it. - An older question that you already answered comes back as plain text rather than as clickable options — answering it twice would apply a stale choice on top of newer ones.
When you select a section in the editor and ask the assistant to connect it to a data source, it no longer asks you which section you meant — selecting it is the answer. **Where:** Site editor → select a section → chat "Connect this section to <URL>" **How to test:** With a section selected, send "Connect this section to https://example.com/data". The assistant should proceed to plan the data connection instead of returning a "which section?" clarification. With nothing selected, the clarification should still appear. **Watch for:** A clarification prompt firing even though a section is clearly selected.
Motion Studio can now caption your own footage: upload a clip, we write out every word with its exact timing, you pick a style, and you get an MP4 with the captions burned in — your original picture and sound untouched. **Where:** Creative Studio → your brand → Motion → "Captions" (or add `?mode=captions` to the Motion page). New videos start from the "New from video" panel on the left. **How to test:** 1. Open Creative Studio, choose a brand, then open Motion. 2. Switch to Captions and upload an MP4, WebM or MOV — up to 80 MB and up to 90 seconds. 3. Wait for "Writing out every word" to finish (usually under a minute). The preview appears with captions playing over your video. 4. Try all six styles — Word pop, Karaoke, Chunks, Lower third, Bold outline, Minimal. Each one re-lays the same words instantly. 5. Turn on "Use brand colour" to recolour the highlight. 6. Fix a misheard word in the Caption lines list and save. Use the arrows on a line, or "Shift all", if a caption lands early or late. 7. Click Export MP4 and wait for the render to land in the list, then download it. **Watch for:** - Captions drifting out of sync toward the end of a longer clip (most likely on footage shot at 29.97 fps). - A caption highlighting the wrong word, especially in the Karaoke style over music or accented speech — lines flagged with a warning triangle are the ones we were least sure about. - Exported audio sounding different from the original — it should be bit-for-bit the same track. - Text sitting under a platform's own UI at the bottom of a vertical video. - Videos longer than 90 seconds or heavier than 80 MB should be refused up front with a clear message, not after a long upload.
Ad and social creatives now look like they were art-directed rather than templated. The image model gets a focused creative brief instead of raw account data (which sometimes leaked business names — even misspelled — into the picture), generation runs on a model that reliably keeps the copy area clean, and the finished composite is craftier: the text band reaches the edge of the frame, button text automatically flips dark on light brand colors, text sitting directly on a photo gets a subtle shadow, your brand's font family is matched, long headlines shrink to fit instead of overflowing, and upscaled images are sharpened. **Where:** Creative Studio → open a business → Create tab → generate an ad or social post **How to test:** 1. Generate an ad for a brand whose name could plausibly appear in a scene (e.g. a marketing SaaS) — the image must contain no text, no brand name, and no gibberish lettering, with the lower part of the frame kept visually quiet for the copy. 2. Let the composite render: the dark band behind the copy should reach the very bottom edge of the image with no strip of raw photo below it. 3. Give the brand a pale accent color (light yellow/mint) and confirm the button label renders dark, not white-on-white. 4. Generate a batch of 3–4 variants from one prompt — they should differ visibly (angle, light, crop), not be near-duplicates. **Watch for:** any generated image containing readable or gibberish text; copy overflowing the image or its band; a button pill wider than the safe area; older saved creatives changing appearance (they must keep their original renders — only newly composed ones use the new rules).
The AI assistant can now remove, switch off, or reconfigure the built-in popups and floating widgets on your site — it used to fail on those requests. **Where:** Site editor → AI assistant (chat or voice) **How to test:** On a site with a newsletter popup, exit popup, announcement bar, WhatsApp bubble, countdown, social-proof toasts, or back-to-top button, ask the assistant "remove the newsletter popup", "turn off the announcement bar for now", "make the newsletter popup show after 10 seconds", or "what popups does my site have?". The change should apply immediately and show in the preview. **Watch for:** Asking about an add-on you never added should get a clear "that isn't on this site, here's what is" answer rather than a silent no-op. Editing an add-on's settings should no longer switch a disabled add-on back on or undo a page-specific placement.
The separate Brand Studio area is gone — editing your brand kit, browsing your saved creatives and assets, and linking the brand to websites now happen right inside the Creative Studio workspace, next to where you create. Old Brand Studio links redirect to the same brand's new home. **Where:** Dashboard → Creative Studio → open a business → Brand Kit / Library / Websites tabs **How to test:** 1. Open a Creative Studio workspace and switch between the Create, Brand Kit, Library and Websites tabs — each should load, and brand kit edits should save and be reflected after switching away and back. 2. Visit an old `/dashboard/brands/{id}?tab=library` URL and confirm it lands on the same brand's Library tab in Creative Studio. 3. Click "New brand from scratch" on the Creative Studio home, create a brand, and confirm you land on its Brand Kit tab. **Watch for:** brand kit edits not persisting; the Websites tab failing to list linked sites; old bookmarked Brand Studio URLs erroring instead of redirecting; the Create tab behaving differently than before (it should be unchanged).
Bulk image uploads with a mix of photo types — portraits, screenshots, product shots — stopped getting their automatic categorization and alt text, and silently kept generic filename-based labels instead. **Where:** Site editor → Pages → Assets → bulk upload a folder of mixed images **How to test:** Bulk-upload a batch that mixes headshots, screenshots and regular photos. Each image should come back with the right category and a real, descriptive alt text — not the filename. **Watch for:** Any image in a bulk-uploaded batch stuck with a generic "gallery" category or filename-derived alt text.
When you bulk-upload a mixed batch of images in chat, the AI vision pass again re-tags them — headshots and screenshots are detected, real descriptions are written, and images are sorted by their best-fit purpose instead of all staying in one group. **Where:** Site editor chat → attach several images → Upload **How to test:** Upload a mix of headshot photos and app/website screenshots with the purpose left on "gallery"; within ~90s the assets pick up `headshot`/`screenshot` tags and real descriptions. **Watch for:** Only newly uploaded batches self-organize — images uploaded before this fix are not retroactively re-tagged.
The background pass that categorizes and labels bulk-uploaded images could be cut off before it finished on larger batches, leaving some images stuck with generic labels instead of their real category and description. **Where:** Site editor → Pages → Assets → bulk upload a folder of images **How to test:** Bulk-upload a larger batch of images (a dozen or more) and confirm every image eventually gets a real category and descriptive alt text, not just the first few. **Watch for:** Images in a large bulk-upload batch left with a generic "gallery" category or filename-derived alt text.
Captions were sometimes tagged as the wrong spoken language, and the word timing could drift noticeably off the audio. Both are fixed — captions now assume English unless told otherwise, and the word-by-word timing is measured against the audio rather than estimated. **Where:** Creative → Motion → Captions, after uploading a video **How to test:** Upload a short clip with a few spoken words and let it transcribe. Check the language shown next to the video size and duration — it should read EN, not a language you did not speak. Play the preview and watch a word-by-word caption style: each word should light up as it is said, not before or after. **Watch for:** A language other than EN on an English clip. Highlights that consistently lead or lag the voice, or that lag more and more as a longer clip plays. Clips over a minute are the best test — drift is easiest to see near the end.
When we pick a photo for a product, menu item, room or listing, we now check that the photo is actually about that item — and leave the slot empty rather than publish something unrelated. **Where:** Site editor → Content → any collection (Products, Menu, Rooms, Properties, Cars…). Applies when a site is generated and when you add a new collection. **How to test:** Generate a site in a catalog industry, or add a collection page, then open Content and review the row images. Items with distinctive brand-style names ("The Signature Supima Crew") should show a relevant photo or a plain placeholder — never an unrelated stock photo. **Watch for:** More placeholder tiles than before on collections where good stock photography genuinely doesn't exist. That is the intended trade: a placeholder you can replace beats a wrong photo you have to notice first. You can drop in your own image on any row from the same screen. Existing sites keep the images they already have — regenerate the collection, or replace a row's image directly, to pick up the new behaviour.
Product and menu tiles now pull stock photos that match what the item actually is, instead of latching onto a stray word in a fancy product name. **Where:** Any catalog page (Shop, Menu, Listings) on newly generated or re-fulfilled sites. **How to test:** Generate a catalog site (e.g. a t-shirt store) and open the Shop page. Each tile's photo should match its category — a "T-Shirts" item shows a t-shirt, not a payment terminal or a tablet. **Watch for:** This is a generation-pipeline improvement, so it applies to newly generated collections. Existing sites keep their current images until the collection is regenerated or its images are re-fulfilled.
A couple of visual concepts (type-led graphics and "printed text in the scene") could still get picked for a post that already has a text overlay, undercutting the earlier fix that keeps generated artwork free of lettering. **Where:** Creative Studio → Social planner → "Create visual", and Brand Studio → Create **How to test:** Plan a text-style post and generate a visual a few times. None of the resulting concepts should be built around in-image lettering when a copy layer is already present. **Watch for:** A concept where legible text is the point of the image — a poster-style graphic, or a sign/screen meant to be read — showing up on a post that also has composited copy.
When you connect a section to an outside data feed, the live items now appear on your preview link right away, so you can confirm the mapping looks right before you publish instead of after. **Where:** Site editor → chat panel → ask the assistant to connect a data source to a section, then open the site's shareable preview link **How to test:** 1. Open a site and ask the assistant to connect an outside feed to a section — for example: "Connect our events feed at <your URL> to the events section." 2. Wait for it to confirm the connection. 3. Open the site's preview link. The section should show the live items from the feed. 4. Publish the site. The published page should show the same items the preview showed. 5. Change something at the source, wait about five minutes, and reload the preview. The updated items should appear. **Watch for:** - Preview showing the old placeholder or hand-written content while the published site shows the live feed — the two should agree. - A section that shows live items in preview but is empty (or shows different items) once published. - Sections you manage through the Collections editor being taken over by an outside feed — those should keep showing your own rows. - A section bound with a specific item limit or field mapping suddenly showing unmapped or too many items.
When you connect a section to a live data source, the editor canvas now shows the real items straight away instead of the placeholder content you started with. **Where:** Site editor → any section you have bound to a data source (via the assistant's "connect a data source", or Settings → Data Sources) **How to test:** Connect a data source to a section, stay in the editor, and look at the canvas. The live rows should appear without publishing and without opening the shareable preview link. Reload the editor and confirm they are still there. **Watch for:** Sections that stay on their placeholder content; live rows appearing on the published site but not in the editor; a section flickering between live and placeholder rows while you type elsewhere on the page.
When you asked the editor to connect a section to an endpoint that replies "401 Unauthorized", it sometimes just told you the address needed a login and stopped — even though it can now collect the key for you. It now opens the secure key box every time. **Where:** Site editor → select a section → chat → "Connect this section to https://your-api.example.com/items" **How to test:** Point a section at an endpoint that requires an API key or bearer token. Instead of a dead end, the "Connect a private data source" box should open so you can paste the key straight into locked storage — never into the chat. **Watch for:** An endpoint that is genuinely unreachable (404, wrong address) should still say so plainly rather than asking you for a key.
Asking the assistant to connect a section to a public JSON/REST feed now connects it instead of failing with an "internal database error". **Where:** Site editor -> Edit mode -> select a section -> chat: "Connect this section to <public API URL>" **How to test:** Point a repeated-card section at a public, credential-free JSON endpoint. The assistant should test the endpoint and bind it, not report an unexplained error. **Watch for:** Any lingering "unavailable" or database-error responses on reachable, no-auth endpoints.
When you ask the site-builder chat to connect a section to a web address or API endpoint, the request now always reaches the data-source connector — which tests the endpoint for real and tells you the truth instead of inventing a fake connection. **Where:** Site editor → refine chat (e.g. "Connect this section to https://example.com/feed.json"). **How to test:** Select a section and ask to connect it to a URL. A reachable public feed gets wired up; an unreachable one is reported as unavailable with no page change; one that needs a key opens the secure credential box. **Watch for:** The chat must never claim "the connection now works" for a URL that was never actually reached, and must never string-swap the URL into the JSX as a cosmetic edit.
When you connect a section to an outside web address, the confirmation now says plainly that the live records appear once the site is published — instead of reading like the page had already changed. The assistant also refuses to connect a section that does not exist on your site, rather than reporting a connection that could never show anything. **Where:** Site editor → select a section with repeating cards → chat panel **How to test:** 1. Open a site in the editor and select a section that shows a list of cards. 2. Ask: "Connect this section to https://jsonplaceholder.typicode.com/photos?_limit=6 while preserving the current layout." 3. The reply should confirm the connection AND say the section keeps showing its current items in the editor until you publish. 4. The editor should not flash or reload as if the page had changed — because it hasn't yet. 5. Open Settings → Data Sources: the new connection is listed against a real section of your site. 6. Publish the site and open the published page — the section should now show the connected records. **Watch for:** - A confirmation that says "Done" with no mention of when the data appears. - A connection listed against a section name that doesn't exist on the site. - The editor reloading the whole design after a connection that changed nothing on screen. - A section that was bound directly (rather than connected) should still update as before.
If your footer linked to a Cookie Policy, GDPR or Accessibility page, publishing used to quietly drop that link from the live site because there was no page behind it. Publishing now builds the missing page for you and keeps the link. **Where:** Site editor → Publish (the live site's footer) **How to test:** Open a site whose footer has a "Cookie Policy", "GDPR" or "Accessibility" link, publish it, then open the live site and click that link — it opens a real page instead of disappearing from the footer. **Watch for:** Pages are only created for links your footer actually has, so a site that never linked to them will not gain new pages. If a compliance link points somewhere we cannot build a page for (for example a custom "Compliance" page), the link is still removed — but the publish now tells you which link it removed so you can add that page yourself.
Editing section text that happened to include certain invisible or malformed characters (sometimes produced by AI rewrites) used to fail to save with an error. Those characters are now automatically cleaned up so the edit saves normally. **Where:** Site editor → any page → editing a section's text content **How to test:** Rewrite a section's copy through the assistant a few times in a row, then save. The save should succeed every time instead of occasionally failing with "Failed to update variation content." **Watch for:** A save still failing on a specific piece of text — that would point to a different unsanitized character making it through.
Asking the assistant to rewrite your page copy without touching the design now saves the new wording — previously it reported success while the site stayed unchanged. **Where:** Site editor → chat panel (any page) **How to test:** Open a site, ask "Rewrite the copy across the page to be punchier, but keep the design, layout, images and colours exactly as they are." Wait for the change to apply, then refresh the page. The new wording should still be there, the design untouched, and a new entry should appear in version history so you can undo it. **Watch for:** Copy reverting to the old wording after a refresh; a change that applies but never shows up in version history (which would leave you unable to undo it).
Asking the chat to rewrite your copy now changes the page you can see, instead of quietly updating a copy of the text behind the scenes. **Where:** Site editor → chat panel, on a site where you've picked a design **How to test:** Open a site with a chosen design, ask the chat to "rewrite the hero headline to be punchier", and watch the preview — the new headline should appear on the page, and still be there after a refresh. **Watch for:** The assistant rewording text that isn't the text on screen; the chat saying it applied a change while the preview stays the same; other parts of the section (button label, image, description) changing when you only asked about the headline.
Asking the assistant to just rewrite wording could, in some cases, also rewrite fields that aren't prose — an email address, phone number, currency symbol or icon name — silently breaking contact links or icons. **Where:** Site editor → AI assistant chat → wording/copy-only rewrite requests **How to test:** Ask the assistant to rewrite a section's copy on a page that includes a phone number, email address or an icon. After the rewrite, the phone/email/icon should be unchanged and still functional; only the prose should change. **Watch for:** A contact link, phone number or icon that stops working after a copy-only rewrite.
Editing an ad or social post no longer lags behind your cursor, and the canvas finally does the things you expect it to: zoom, edit text in place, add your own images on top, and drop a shadow behind copy that is fighting a busy photo. **What changed** - **Dragging is smooth.** Moving a block used to re-render the whole design on every mouse movement. Now the block follows your pointer directly and the design is saved once when you let go. - **Zoom and full canvas.** Zoom in and out, snap back to fit, or open the creative full-screen with the controls above the preview. - **Double-click to edit text.** Double-click the headline, offer or button on the canvas and type. The fields below still work exactly as before. - **Resize by dragging a corner.** Corner handles appear on whatever is selected — images and the logo resize, text gets bigger or smaller. - **Add your own images.** "Add image" places any picture from your library on top of the creative: a second logo, a product cut-out, a badge. Move it, resize it, set its opacity and corner rounding, and order it against the other layers. - **Drop shadow for text.** Each of the headline, offer and button can carry its own shadow, with colour, strength, softness and drop. - **Arrow keys nudge.** Select something and nudge it a pixel at a time, or ten with Shift held. **Where:** Creative Studio → open a finished ad or social post → Edit layout **How to test:** Drag the headline across the canvas — it should track your cursor with no stutter, and the snap guides should still appear at the margins and centre line. Zoom to 200% and drag again. Double-click the headline, change the wording, and click away. Add an image, drag a corner to resize it, and move it below the headline in the layer list. Turn on the headline's drop shadow. Then apply the copy and confirm the downloaded creative matches the canvas exactly. **Watch for:** The rendered result must match the canvas pixel for pixel — if a block sits in a different place in the final image than it did in the editor, that is the thing to report. Images you place come from your own library; if one is missing from the picker, check it is not a site screenshot or a chat upload, which are deliberately excluded.
Creating an ad or a social post starts with suggestions written for your business — pick one and go straight to review, or just describe what you want in your own words. **Where:** Creative Studio → open a workspace (Dashboard → Creative → pick a business) **How to test:** 1. Open a Creative Studio workspace. Instead of the "What are you creating?" picker you now land on "What should we make?", with a short loading state and then three or four idea cards. 2. Each card should be specific to that business — its own goal, its own format, its own words — not generic marketing copy. Cards may show one of your own library images if a good one exists. 3. Click a card. You should land straight in the Review step with the format, campaign goal, visual direction, copy and reference images already filled in, and the credit cost shown before you generate. Nothing is charged until you press Generate there. 4. Type something into the chat box instead ("an Instagram post about our autumn menu"). You should get back either a proposal card with "Use this concept" and "Adjust details", or one short question with two to four answers to click. Clicking an answer should carry the conversation forward, never restart it. 5. Click "Use the detailed form instead" at the bottom. The original step-by-step wizard should open exactly as before, and still work end to end. 6. Opening Creative Studio from a planned social post or a site launch should still go straight to the configure form with the prompt already written. **Watch for:** ideas that mention a price, discount or guarantee the business never supplied; a card that lands in Review with an empty format or missing copy; the "Adjust details" path losing what was already decided; the detailed form behaving any differently than it used to; and Brand Studio's own Create tab, which should be completely unchanged.
The "Reference assets" grid in Creative Studio (ads and social posts) was showing system captures — site preview screenshots, images pasted into design chat, and favicons — and could silently pre-select them as generation references. These are now excluded, selected references show a clear check badge with a selected count, and a "Clear selection" control lets you generate from the brand kit alone. **Where:** Creative Studio → pick a website → Ad or Social post → Configure step → Reference assets **How to test:** Open Creative Studio for a site that has a design-chat image upload or preview screenshot; confirm it no longer appears in the reference grid. Select and deselect a real library image; the check badge and "N selected" count should update, and "Clear selection" should empty it. **Watch for:** Sites whose only images were system captures now show the "No website images are available" note instead of a grid — that is intended.
Creative Studio is now a workspace instead of four separate screens — with a home that shows every video, visual and plan you have made, a real content calendar, and the ability to brand a video you shot yourself without adding captions. **Where:** Dashboard → Creative Studio **How to test:** 1. Open Creative Studio. Pick a job from the four cards (Create a visual / Plan content / Video / Design a logo) — you are asked which website it is for *after* choosing, not before. 2. Check "Continue where you left off": every visual, video, uploaded clip, plan and logo you have made should be there, across all your brands. Click one — it reopens in its own editor. 3. Inside a brand, use the left rail to move between Visuals, Calendar, Video, Logo, Library and Brand Kit. The website you picked stays with you the whole way. 4. Open Calendar. Generate a plan, then drag a post to another day and edit its caption in the side panel. Click "Create the visual" on a post — when the image finishes it appears back on that post. 5. Add a voice under Brand Kit → Voices, then generate a plan as that voice and filter the calendar by it. 6. Open "Calendar — all brands" from the home page to see every client's week on one grid. 7. Open Video → "Upload & edit your footage". Upload a clip and click **Suggest graphics** without transcribing it first — you should get a title card, logo and end card on your own footage. Then add captions separately if you want them. 8. Export the result and check the MP4 has your graphics burned in. **Watch for:** old links should still work — bookmarks to the Social Planner, Motion Video (including the captions mode) and the brand workspace tabs all redirect to the right place, carrying the website they were opened for. A plan you made before today should appear on the calendar the first time you open it.
Sections you create with your own prompt now follow your site's existing colours, fonts and copy instead of arriving in a generic style you have to fix afterwards. **Where:** Site editor → select a section → Section Explorer → Library tab → "Create Custom" **How to test:** Open a site with a distinctive colour scheme, describe a new section (for example "a three-column list of what we offer"), and generate it. The result should already use your site's palette and typography, and reuse your real business copy where it fits — not placeholder text in default styling. **Watch for:** A generated section that still comes back in generic greys/blues, or one that invents business details instead of using the content already on your site.
Content and blueprint edits made during the planning stage — before any design has been generated — were sometimes silently dropped from version history, making them impossible to undo. **Where:** Site editor → any site still in the planning stage (content plan, blueprint chat) before the first design is generated **How to test:** On a brand-new site with no generated design yet, make an edit through blueprint chat or the content plan, then open version history. The edit should appear as a restorable version. **Watch for:** A planning-stage edit that doesn't show up in version history, or an "undo" that has nothing to revert to.
Content collection rows now show a clear "Edit" button, so you can open any item to add an image or edit its details — no hidden gestures required. **Where:** Site editor → Content / Collections → open any collection type's items table. **How to test:** Import a CSV (or open an existing collection). Each row now has an "Edit" button in a new Actions column; click it to open the full item editor. Clicking an empty Image or Description cell also opens the editor. Double-clicking the title still works as a shortcut. **Watch for:** Rows with no Edit button, an Edit button that doesn't open the editor, or misaligned columns on wide collection types (e.g. Products) — the table scrolls horizontally rather than breaking.
Previews on your site's own web address could fail with "Preview Error — Failed to fetch dynamically imported module" when the page used certain building blocks (payments, some interactive components). The page is now able to load those pieces correctly. **Where:** Your site's preview link (e.g. `your-site.fullyloadedsites.com/preview`), and any preview opened on a connected custom domain. **How to test:** Open a site preview on its own web address and confirm the page renders instead of the red error screen. If you saw the error before, reload once after this update. **Watch for:** A preview that still shows the red error screen and names a different address — that would be an unrelated failure, so send us the full message.
Generated footer and header menus sometimes listed an item — "Categories", "Careers", "About" — pointing at a page the site never got, so clicking it landed visitors on the not-found screen. Those items are now removed while the site is being built, and shortcut links like "Help Center" that jump to a section of the homepage now work from every page instead of only from the homepage itself. **Where:** Any generated site — the footer (and header menu) on every page. **How to test:** Generate a site, open the editor preview, and click every footer link from an inner page such as About. Each one should open a real page or scroll to the right homepage section. **Watch for:** A footer column that legitimately shrinks (its dead entries were dropped). Privacy Policy and Terms links must always survive — those pages are created at publish time.
Buttons like "Shop the Collection" on a newly generated inner page could link to a made-up address (for example `/shop` on a store whose catalog actually lives at `/products`), sending visitors to a not-found screen. Those destinations are now checked against the pages the site really has and repaired automatically. **Where:** Any newly generated site — inner pages such as About, plus the buttons inside them. **How to test:** Generate a site with an About page, open it, and click every call-to-action button. Each one should land on a real page or scroll to the matching homepage section, never on a "no page at …" screen. **Watch for:** A CTA that jumps to the homepage and scrolls — that is the intended repair when the button named a section rather than a page. Links to pages that are created later at publish time (privacy policy, terms) are deliberately left alone.
When you upload a batch of images and name them (for example changing "alice-headshot_final" to "Alice"), that name is now what you see in your site's asset library — previously the library kept showing the raw file name and your label was nowhere to be found. **Where:** Site editor → the "..." menu → Assets (or Mission Control → Manage Assets) **How to test:** Attach 8 or more images in the editor chat to open the bulk upload dialog, rename one image to something meaningful, pick a purpose, and upload. Open the asset library — the card should be titled with the name you typed, and searching for that name should find it. **Watch for:** Images you did not rename should fall back to a tidied-up version of their file name (for example "team photo 02"), never a blank title. Existing assets keep whatever name they already had, and you can still rename any asset inline.
Motion Studio can now make videos up to three minutes long — planned as a set of scenes, generated one at a time, and exported as one continuous file. **Where:** Creative Studio → Motion Video → New video → pick the "Long (up to 3 min)" length. Standard-length videos with five or more text beats are built the same way automatically. **How to test:** 1. Open Creative Studio for a brand and start a new video. 2. Choose the Long length, fill in a message and a few text beats, and generate. 3. Watch the progress panel — it names the scene it is on ("Scene 2 of 6"). 4. When it finishes, use the Scenes strip under the player to click through the cuts. The selected scene plays above, and "Change something" applies to that scene. 5. Press "Regenerate scene" on one card — only that scene is rebuilt, and only that scene is charged. 6. Export MP4. The finished file plays all the scenes back to back. **Watch for:** scenes that don't look like they belong to the same video (the palette and typefaces of the first scene are meant to carry through every later one); a scene that fades to black before the next one starts; an exported file that is shorter than the length shown on the project; GIF being offered on a multi-scene video (it should be greyed out — export MP4 instead); and a long export sitting in "rendering" for more than about half an hour.
A video composition that included a standard font-preconnect tag was incorrectly flagged and silently forced into an extra repair pass, adding time to generation for no reason. Simple video briefs also now run at a lighter setting that finishes noticeably faster without giving up quality. **Where:** Creative Studio → Motion → generate a video **How to test:** Generate a short, simple video brief — it should complete faster than before with no unnecessary retry. **Watch for:** Generation taking longer than expected on a simple brief, or a composition rejected purely for including a font-preload link.
If part of a video generation ran slow and had to fall back to a backup step, a follow-up repair pass could re-walk the same slow path and run out of time, ending the generation with no clear error instead of a usable result or a clean failure message. **Where:** Creative Studio → Motion → generate or refine a video **How to test:** Generate a video and watch it through to completion, including any repair/retry steps — it should either finish or fail with a clear, readable error, never hang or die without explanation. **Watch for:** A generation that appears to stop responding with no progress update and no error shown.