Changelog

Support answers now point at the control instead of just describing it. When the VidNext Assistant (or a person from the team) mentions a setting, the reply carries a Show me chip next to it — click the chip and the app opens the right page, scrolls to that control and highlights it, opening a collapsed section first when needed. The help panel folds into a Back to chat pill while you look, so one more click brings the conversation back with your draft intact. If the control isn't on the page yet — a step that only appears after you enter a topic or pick a voice, for example — the guide points at what to do first and continues to the control on its own once it appears. The same Show me in the app links now sit next to the steps in these docs, and the ones you receive by email open the app in the same way. Details in Help & Support.

You can now talk to the VidNext team without leaving the app. A life-buoy Help button next to the notification bell opens a side panel on every page with your conversations, a new-conversation form and the chat itself. Get help on a video that stopped (dashboard card, project page, auto view) opens the form with that video, its status and its error already attached, and the Page & browser details card shows exactly what else we attach so you can untick it. Paste screenshots straight into a message, or drop files — up to 5 per message, 25 MB each.

Right after you send, the VidNext Assistant may answer — built from these docs, clearly labelled as automatic, with This solved it and I still need a person underneath — and you can keep chatting with it. When the docs don't cover a question it asks whether to pass it to a person instead of guessing — Yes, connect me with a person alerts the team, No, that's all closes it quietly — pricing and credit questions are answered from the docs, and anything about your own payments, refunds or account goes straight to the team. The header shows whether the team is online and how fast it usually replies, replies appear in the open conversation within a second or two, and you can turn on browser notifications for them. Replies from the team arrive in the app (bell notification, a New marker in the thread) and by email, and answering the email lands your reply in the conversation. Conversations move through Open, Waiting for you, Resolved and Closed; mark one as solved yourself, rate it once it is resolved, or reopen it by replying. A new Support replies toggle in Settings → Notifications controls the emails. Everything is described in Help & Support.

New to VidNext? A Watch the demo button now sits under the heading of the create page and next to the YouTube guide on the dashboard. It plays a short walkthrough of making a video, and switches to a music-video demo when your narration source is Music video. The button appears once a demo is available for that mode.

The slider on the pricing page now ends in a grab handle with left/right arrows, and a Drag to size your pack hint sits under it until you move it for the first time. Drag the handle (or use the arrow keys once it has focus) to move between the twelve packs; the video count, price, and per-minute rate update as you go. See Pricing & plans.

Partner promo codes now work on the pricing page: a code you arrive with through a partner's link is picked up automatically, or enter one under Have a promo code? at the bottom of the page. The discounted prices are shown before you pay, the discount is applied at checkout, and the payment confirmation names the code. Most codes are for your first top-up only, and discounts never stack — you always get the best one. See Pricing & plans.

Story images now always appear as your own picture in the image design you chose for it (or one of your enabled image designs). Previously an image the script did not reach could be added as an extra scene with an empty or unrelated design, and every later scene could drift out of sync with the narration. Images the script does not place itself now take over an existing scene near their intended point, so the video keeps its length, its order and its narration.

Web image selection now judges every candidate against the video's topic, era, region and language, not only the scene's keywords: a picture that contradicts them (a different country's living room in a village documentary, a modern colour photo in an archival story) is not used. Watermarked pictures, stills with burned-in captions in another language or a broadcaster logo, and photos of famine, war dead or disasters used to decorate an unrelated point are rejected as well. Cards that name a specific person, brand or product model search for that exact name.

Music videos can now carry the graphics that make a clip look produced, all chosen in the music video card before you generate:

• Equalizer — 24 visualizer styles in four families (bars, rings, waveforms, fields) that move with the real beat and spectrum of your song. Choose the placement, a colour mode (style-pack accent, custom colour, white or a spectral ramp), intensity and glow.
• Lyrics — 21 big, animated karaoke styles synced to the singing word by word. The kinetic family (Kinetic Cascade, Kinetic Slam, Typo Poster) builds each line word by word with a different move per word and sizes that follow the singing; Infinite Depth keeps lines travelling away into a tunnel; Drift & Fade and Split Reveal are cinematic reveals; the karaoke pill, highlight sweep, gradient, neon, glitch, typewriter and knockout looks complete the set. Size, font and colour are yours, the position follows the style unless you pin it; lyrics replace captions for that video. The transcript is cleaned up and re-synced to the song before rendering — review the lyrics first for the best result.
• Title card — the song title and artist animated before the first line, in four variants.
• Live preview — the Overlays block previews every enabled overlay together over sample footage while you adjust it, with pause, restart, scrub and a larger view; the equalizer gained an opacity control.
• Editor — the equalizer and lyrics elements are editable on the Overlays track: style, colours, placement, intensity, glow and opacity for the equalizer; style, position, size, font and a Lyrics & timing list for the lyrics, where every word or line and its start and end can be corrected, split, joined or set at the playhead. Exports use the edited words and times.
• Lyrics review — music videos now review their lyrics in a dedicated dialog: timed lines with their moment in the song, in-place edits, Enter to split and Backspace to join lines. Edited words are re-synced to the singing before rendering; there is no proofreading step for lyrics.
• Vary by song section switches an overlay to a companion style on the chorus and calms it down on the intro and outro; the Footage dim slider keeps everything legible over bright clips.
• Every style has a live preview over sample footage in your format, and all three overlays are editable later in the editor (style, colours, placement, dim).

Details in Music videos.

Map scenes that showed two places close to each other — a route between neighbouring towns, a cluster of ports, a city with its suburbs — could print both names on top of each other, leaving an unreadable smudge next to the marker. Fixed:

• Names never overprint. Every map style (routes, spread, zoom with secondary markers, arrows) now places each name in a free spot around its marker: the other side, above, below, or one row further out with a thin pointer line back to the marker. Names that were already legible stay exactly where they were.
• Short routes get a real map. A route whose stops sit a few dozen kilometres apart used to be framed like an intercontinental flight, with both markers on one spot. The camera now dives to the scale of the route, so a county-level evacuation route shows two distinct markers and the path between them.

The footage check introduced on 2 September was rejecting far more than it should and then searching for hours to replace what it had thrown away. Fixed:

• Any moment of the named film counts. A clip from the right film at a different moment than the one described is no longer marked "off-target" — it is kept, and only footage from a different film, a presenter or reaction video, a live performance, a themed event or a collage is rejected.
• One search pass per scene, not one per keyword. A scene that still needs footage after the check is re-searched once, from its film's own moments first, in a single analysis instead of dozens; a second pass runs only when the first found nothing usable. The whole re-search is capped per video, so a long song can no longer spend an hour on it.
• Scenes with enough footage are left alone. A scene that already has enough distinct footage but fewer clips than beats replays its clips and borrows footage the viewer has not seen yet instead of being re-searched.
• Stock sections are never re-searched on YouTube. They are chosen by the stock library's own scoring and were previously sent through the film re-search for nothing.
• Fewer junk sources up front. Search results for film scenes are now screened by title and thumbnail before the footage is analyzed, so nature clips found for "Before Sunrise" or a dance show found for "Moulin Rouge" no longer reach the check in the first place.

Music videos of the same length are back to their previous cost and generation time. Details in Music videos.

Music videos were repeating the same shots — a five-beat scene often had only two clips to cut from — and the stock check could take longer than the rest of the video. Changed:

• One distinct shot per beat. Scenes from the same film (or the same passage of the song) now share one footage pool, and VidNext asks for as many different moments as the scene has beats, plus spares. A film whose pool still comes up short is searched once more from its generic scenes.
• Cuts on real shot boundaries. Each clip's shot changes are detected in the footage itself, so a beat never starts mid-shot and a long clip with several shots gives several cuts. Consecutive beats alternate between wide, medium and close shots when the footage allows.
• Same film first. A scene that runs short borrows unseen footage from its own film before anything else; borrowing from other scenes is the last step, and the quality report now flags any scene where footage still had to repeat.
• Sharper source picks. Uploads below 720p are skipped while HD ones are available, clips with a visible channel watermark are kept only as a last resort, and long uploads are sampled rather than watched end to end.
• Faster stock check, for every video. Stock clips are screened from their preview frames first; only the best few per scene are downloaded and scored as video. This applies to stock sections in narrated videos as well, including videos with a single stock scene.
• Cuts on the downbeat. The song's bars are now read from the audio (accents and chord changes decide where beat one is), so cuts land on bar lines instead of on whichever beat a section happened to start on — the montage finally moves with the music rather than merely at its tempo.
• Song language is auto-detected. The language pill in Music video mode now reads Auto-detect by default and is labelled as the song's language. It used to carry the app language silently, and a wrong forced language made the lyrics come back translated instead of transcribed.
• Fewer re-searches. A clip with a channel watermark or a shot from another film in your brief no longer counts as missing footage, so scenes that already had usable material are not re-searched.

Details in Music videos.

Starting a video from a planner idea could occasionally create the same project twice — a double-click on Start now, or a second click on a card that hadn't refreshed yet, each started a full generation, and the extra one later appeared as a duplicate card. Fixed:

• One start per idea. Once an idea has a video in progress, a second start is refused with a clear message and the board refreshes to show the running project. The idea can be started again after that video has finished, failed or been cancelled.
• The Start button tells the truth. An idea whose video is still in progress shows the progress link instead of a Start button — in every column, including an idea moved back to Ideas while its video is generating.
• Faster feedback. Start now shows its spinner on the first click, and the dialog can't be dismissed while the start is in flight.
• No stranded credits. If a start fails before generation actually begins, the held credits are released immediately.

This applies to every way of starting a planned idea: the board card, the idea details, the calendar's Needs video panel and Customize settings. Details in Channel Planner.

Footage search can now locate the moment a scene needs by reading a candidate video's transcript first and only then looking at the frames around it, instead of scanning every candidate from start to finish. Each scene group is checked against up to ten candidate videos at once, and sources up to 60 minutes long are accepted, so full episodes and long documentaries are no longer skipped in favour of short compilations.

Every clip that reaches your project was visually verified: a candidate that was only described in speech but never looked at is rejected and searched again. The change is rolling out gradually to accounts; nothing changes in how you create a video.

Music video mode now vets every downloaded clip before it reaches the edit. Footage that is a still image or poster, a title card or credits, near-black, or clearly from a different film than the one requested is rejected, and the scene searches again — first the film's own moments, then the other films in your brief, then cinematic footage matching the song's mood and motif. Clips with burned-in subtitles or a sub-HD source are kept only as a scene's own last resort when nothing cleaner exists.

Within a scene the same seconds of a clip are no longer shown twice: a scene with fewer clips than beats now plays different moments of its clips instead of repeating one window, and a scene that ends up with no footage of its own borrows clips the viewer has not seen yet before repeating anything. Search results from auto-generated "Topic" music channels, trailers, soundtrack uploads and fan tributes are skipped up front.

The create step for music videos is simpler. After choosing Music video, the card asks for the song first, then where the footage should come from — Mixed (VidNext decides per section, the default), One film or show, Several films, Stock & abstract or Nature — and then an optional brief. One line says what you get: real clips cut on the beat, not an AI-generated video. Controls that don't apply to a music video are gone from this mode: the script and attachment menu, sound effects and the narration starters. The script-review switch now reads Review lyrics before generating, since the transcribed lyrics are what it reviews.

Scene transitions in music videos now follow the song: entering a chorus or the outro always gets a transition, other section changes usually do, and beat cuts inside a section stay hard cuts. Music videos start with a Music video transition set (dissolve, fade to black, film burn, flash, impact cut, whip pan), also available as a preset under Scene Transitions. The footage choice presets the Content Mix (Video Clips from YouTube versus Stock Video from the stock library) and the Footage Sources list, both still adjustable, and narration features such as chapter headings, text overlays and calls to action no longer appear in this mode.

Stock footage selection is working again. Since mid-August the stock library clips were found but never scored, so every stock scene quietly fell back to an image or, in a stock-only music video, to a black frame. Music videos now also keep moving footage as a last resort when no clip clears the relevance bar, a video with nothing on screen is refused before rendering instead of being delivered, and the final check samples frames so an all-black render can no longer pass.

Music video mode and Upload my narration find real footage — film and show scenes, stock clips — and cut it to your audio; they can't render animation, invented characters or camera directions. If your brief reads like a prompt for a video generator, the Brief check now explains this before the video starts and offers three real-footage briefs on the same idea, in your language. Pick one, edit your brief, or Create anyway to continue with real footage. A brief that can only be fulfilled with generated footage no longer turns into a video by accident: without that confirmation the project stops before generation and no credits are charged.

Narration uploads are now checked for what they sound like. A music track uploaded in Upload my narration, or a spoken recording uploaded in Music video mode, gets a note next to the file and a dialog on Generate that offers to switch to the matching mode with the file kept. The check is a heuristic — singing, poetry or narration over a music bed can fool it — so your choice always wins.

In Music video mode the Story images and Video reference attachments are hidden: the song and the brief drive the visuals, and neither attachment was used there.

The narration Speed setting (Slowest–Fastest) is now part of your saved presets. Save a preset with a speed picked and it comes back exactly when you apply the preset — including planner one-click videos that use it. Previously the selector reset to Normal every time, and changing the speed now also counts as a preset change you can save back.

Generating a video straight from a planner idea no longer needs any preset setup. Before, one-click generation required the preset's prompt to contain a {{Main Idea}} placeholder — without it the start was blocked, and an idea's description and brief never made it into the video. Now it just works:

• Any preset, zero setup. The first time a preset is used from the planner, VidNext prepares a planner version of its prompt automatically — the topic becomes a placeholder, every style and format instruction stays intact. From then on each planned idea slots straight in.
• The whole idea travels. The prompt now carries the idea's full details — for AI ideas the brief (hook, angle, key points, must-cover facts, keywords), and for manually added ideas their description, notes, tags and thumbnail idea. Manually written ideas used to contribute nothing but a title.
• See it before you start. The generate dialog gained a What we'll generate preview showing the exact prompt, and Customize settings opens the create page with that prompt filled in and editable — the video uses exactly the text you leave there.
• You stay in control. The channel editor's new Planner prompt panel (under the preset picker) lets you view, edit or reset the planner version of the prompt. Your saved preset prompt is untouched — on the create page it loads exactly as you wrote it.

Placeholders are now optional everywhere, and {{Title}} and {{Description}} join {{Main Idea}} for exact placement when you want it. Details in Channel Planner.

Fixed a bug where auto-generated thumbnails could render your whole video brief as the on-image headline instead of a short punchy title — and ignore a detailed thumbnail instruction you'd written. Detailed instructions with exact headline text (colors, layout, a reference photo) now carry through reliably, and headlines always stay short even when generation has to fall back to a simpler layout.

Thumbnail headlines now also follow your video's output language automatically. If your video is set to Ukrainian, Japanese, or any other language, the generated headline and its variants come out in that language — you don't need to ask for it. An instruction that specifies exact headline text is still used verbatim, in whatever language you wrote it.

Thumbnail instructions now also work everywhere, not just in Thumb Lab:

• The project Thumbnails tab gained the "Use my prompt exactly" switch and, with references attached, the extra-web-references switch — the same controls the Thumb Lab composer has.
• The create page's thumbnail popover now takes a custom instruction, up to 3 reference images, and "Use my prompt exactly" — applied when thumbnails generate automatically with the video.

Details in Thumbnails.

The Video Stats Monitor now records a point only when a video actually moves — views, likes, comments, the title or the thumbnail. A flat stretch has no points in it and the chart draws straight through, so the history reads as changes instead of as a wall of identical measurements. Nothing you could see before is gone: every video's current numbers, its title and thumbnail history, and the views-per-hour ranking are unchanged.

The last 30 days keep every collection. Beyond that, each video keeps one point per day — enough to see a trend on an older video without carrying hourly detail forever.

Videos tracked on the Stats page was counting measurements rather than videos, so the number was far larger than the number of channels and videos you actually follow. It now counts tracked videos, which is why it drops sharply.

Collection is smarter about where the detail matters, too: the newest videos on each channel keep being checked around the clock — the launch window where hourly numbers are worth watching — while the full channel refreshes once a day. If you're away from VidNext for a couple of weeks, collection eases off for your channels and catches itself up the moment you're back; Collect now still runs a full refresh on the spot.

Details in Channel Planner.

New ideas on the Board, the channel page and the calendar's ⋯ menu now opens the Idea Studio. Pick a channel, how many ideas you want (the studio suggests a number from the empty slots in your calendar), the horizon, the evergreen-to-trending mix, the formats, and what the AI should look at — your channel's performance, your competitors' latest uploads, live YouTube trends, and web research with the seasonal calendar. The credit estimate is on the button before you run, the run works in the background with live progress, and the bell tells you when the ideas are ready.

Every idea comes as a card with a hook, format and length, a "why now" backed by links, a score, a publish-by date when it's timely, tags, a thumbnail concept and alternative titles. Approve, Decline (with a reason) or save it for Later — with A, D and L on the keyboard — approve everything above a score in one click, and Generate more continues from your decisions; a declined idea never comes back for that channel. Then Add to Ideas, or Add & schedule to drop them straight into the channel's empty slots, timely ones before their date. Each idea keeps its brief, and it's used when you generate the video.

Switch on Automatic ideas per channel — daily or weekly — and fresh ideas arrive as pending, with a Review count on the New ideas button, on the channel Overview and in the Monday summary. The old Add idea and Bulk add forms live on as the Write one and Paste a list tabs.

Details in Channel Planner.

The Planner tab is now a calendar across all your channels: every scheduled post is an event in its channel's colour, Month, Week and Agenda views, and a panel on every post with the posting checklist, the YouTube kit and Mark posted. Drag a post to another day to reschedule it, and drag finished videos from the Ready tray onto a day to schedule them. Posting times are set in each channel's own timezone and shown in yours.

Add channel is a full channel editor: paste a YouTube URL or @handle and Analyze with AI to fill the profile — audience, countries, content pillars, keywords, goals and the AI's insights — then Find with AI to add competitor channels. Every competitor gets a role relative to your channel — Rival (with a 0–100 beatability score: Beatable, Catchable or Uphill), Learn from or Benchmark — and the panel opens with a Your position line so you tick channels you can actually race, not the giants you should only study. Brand-new channels are covered too: with no public subscriber count the planner assumes 0, the rival band always reaches 10,000 subscribers, and the posting timezone follows the audience (a US audience gets New York, not your browser's zone). Tracked competitors are checked the moment you add them, so Recent uploads from tracked competitors on the channel's Overview fills within seconds (and has a Refresh button plus the time of the last check). Each channel page now has Overview, Calendar, Videos, Pipeline and Profile tabs. The editor's Basics section also holds a Default thumbnail style next to the default video preset: projects created from the channel start with that style (leave it empty to pick one per project).

You can also get the plan by email: Email me my plan in the toolbar (or Settings → Notifications → Planner emails) switches on a daily posting plan at the time you choose, a reminder an hour before each post and a Monday weekly summary — and Subscribe gives you a private link that puts your posting plan into Google, Apple or Outlook Calendar.

Details in Channel Planner.

Retry on a variant in a project's Thumbnails tab now re-uses the reference images you attached when you generated it. Before, a retry re-rolled the variant from the style alone and quietly dropped your references, so the new image often looked nothing like the one you were trying to improve. It also keeps the original run's choice about whether to search the web for extra references, so a retry reproduces the same recipe. Details in Thumbnails.

Thumbnail generation is also quicker to fall back when Google's servers are busy: one step on the thumbnail path could previously spend up to five minutes retrying before moving on, and now gives up after 30 seconds like every other step.

The Storage page (account menu → Storage) now works on several projects in one go. Tick the rows you want — or use Select all, Failed, Older than N days and All but newest N — and the bar under the list shows the count and size of your selection with Clean up selected and Delete selected. Both ask once, then show a progress counter and end with a summary of how much space was freed. Details in Storage.

The batch clean-up that used to live in the dashboard's storage panel has moved here and been reworked along the way: large batches no longer fail part-way with a network error, deleting a project that was still generating now refunds the credits held for it, and the storage figures update right after a clean-up instead of a few minutes later.

Deleting a project from the dashboard now removes its card the moment you confirm. Previously the card stayed in place until all of the project's files had been cleaned up — for a large project that could take half a minute — so it looked like nothing had happened, and trying again on a project that was already gone showed "Failed to delete project".

Now a repeat attempt on an already-deleted project simply completes, and your storage usage updates as soon as the project is removed. If a delete does fail, the card comes back and the message says so.

VidNext makes narrated videos about a topic — it doesn't render a shot you describe, the way text-to-video tools like Veo or Sora do. Prompts written for those tools ("static cinematic shot of a cabin in heavy rain, photorealistic, NO camera movement…") used to go straight into production and come back as something their author never wanted.

Now the create page catches this before generation. If your prompt reads like a shot description, a Prompt check dialog explains what VidNext actually produces and offers three topic prompts built from your own idea — one click puts one into the prompt field, and the quick-start format chips are there too. Prefer your original text? Create anyway continues exactly as typed; the check never blocks you.

A live hint under the prompt field also appears as soon as your text looks like a text-to-video prompt, so you don't have to reach the Create button to find out.

Details in Scripts & prompting.

The Terms, Refund Policy, Privacy and Contacts links at the bottom of the account menu now open the page in the language you're using: the English interface links to the English documents, the Ukrainian interface to the Ukrainian ones. Previously the English menu could open the Ukrainian pages.

The editor now has a lite mode for slower laptops. It switches off the parts of the timeline that cost the most on weak hardware — clip thumbnails, audio waveforms, and the preview's pre-loading of upcoming clips — while editing, playback and export stay exactly the same.

The editor decides by itself: a quick speed check when it opens gives fast machines the full editor immediately, slow ones lite mode (a yellow notice with a Turn off button appears), and in-between machines start light and switch to the full editor once the page is running smoothly. A page that stalls for several seconds drops back to lite mode and remembers it. You can also toggle it any time with the ⚡ button at the right end of the playback bar. Your choice is remembered on that device. If a computer can't open the editor at all, add ?lite=1 to the project's editor address once. Details in Editor tour.

Also in this update: scrubbing the timeline while zoomed far out no longer makes the preview reload dozens of clips per step, thumbnails are loaded only for clips that stay in view, and the editor logs [perf:editor] diagnostics in the browser console when something is slow — if the editor ever feels stuck, a screenshot of that console helps us fix it.

Every email VidNext sends about your videos and credits is now a toggle — including two new ones: a reminder when a finished video sits undownloaded for a couple of days, and a short guides series after your first export (prompting, presets, footage control, Shorts, voice and music, redo, the publishing kit). Open the account menu and pick Notifications (or use Notification Settings at the bottom of the bell menu) to switch Video updates, Problems, Credit balance and Tips & product news on or off individually, or pause them all with one master switch — changes save instantly. Purchase receipts, refund confirmations and team invitations are always sent.

The emails themselves now name a video by its generated YouTube title, or simply say "your video" when no title has been generated yet, instead of repeating your full prompt — the Open the project button takes you to the right place either way. Service notices link to your notification settings in the footer; the Unsubscribe link in tips and news emails keeps doing exactly that. Details in Email notifications.

The editor's bottom timeline panel now has a drag handle. Grab the thin divider just above the playback controls and pull it up to see more tracks at once, or down to show as few as two tracks and give the video preview more space. Double-click the divider to go back to the default six-track height; with the divider focused, ↑/↓ steps the height one track at a time. Your chosen height is remembered on that device. See Timeline editing.

The editor is also noticeably lighter on long projects: audio waveforms are now built once per file instead of once per narration segment, clip thumbnails are loaded only for the part of the timeline you're looking at, and the preview does far less work per frame — long videos that used to make the tab unresponsive on slower machines open and scrub smoothly.

vidnext.ai/blog has grown into a full library over the last two months, across four areas:

• Tutorials — your first video step by step, prompts that produce good scripts, fixing one scene without regenerating the whole video, choosing a narration voice, copying a creator's writing rhythm with a video reference, and the thumbnail formula.
• Niches — true crime and history production, finance and tech, and the seven faceless niches worth starting in 2026.
• AI & YouTube — Shorts against long-form, where mid-roll ads belong, running three channels without burning out, and getting monetized before the 1 February 2027 threshold change.
• Product — what a 10-minute video actually costs, itemised credit by credit.

Several of them are built on one video we timed end to end and billed line by line, so the numbers are measured rather than estimated — including where every one of its 442 credits went.

The counters under the dashboard welcome title now tell the whole story. In Progress counts only projects that are actually generating or rendering — your parked drafts have their own Draft counter. Failed counts real failures only; projects you cancelled yourself are listed under a new Cancelled counter instead of being mixed in with errors.

All six counters still work as one-click filters for the project list, and the compact filter row next to the search box has matching buttons for drafts (✎) and cancelled projects (⊘).

Every finished thumbnail variant — in Thumb Lab and on a project's Thumbnails tab — now has an Edit button. Describe the change in plain words ("make the headline bigger and yellow", "remove the person on the right"), optionally attach up to 2 reference images, and confirm. Your words are sent to the image model exactly as written.

The result appears as a new variant right next to the original, marked Edit, with the usual download / analyze / save-style actions — and it can be edited again. The original stays as it was. It costs the same as generating one variant, and the price is shown before you confirm; on the project tab, Retry on an edited variant re-runs that same edit.

Project thumbnails opened from Thumb Lab's Recents also gained a Use in project button, so you can pick the project's thumbnail — including a freshly edited one — without leaving Thumb Lab. And Generate more there no longer runs blind: it opens a dialog prefilled with the latest variant's final prompt and reference images, where you can tweak the prompt, swap references and choose how many variants to make.

If you use the Channel Planner, the Recents panel in Thumb Lab now has a channel selector at the top: see only the thumbnails made for one channel, or all of them. While a channel is selected, new thumbnails you create are saved to it — the composer shows the channel as a chip you can clear.

Project cards on the dashboard and in the Channel Planner now show a still frame of the finished video (or your generated thumbnail) instead of loading a video player for every card. Nothing streams in the background any more: on desktop, hovering a card previews the footage as before; on phones and tablets a tap takes you straight to the project.

The page also stops refreshing while its tab is in the background and picks up again the moment you return, and the moving ambient lights on the dashboard stay still on touch devices. Phones that used to run warm on the dashboard should feel noticeably cooler and faster.

The Cartesia voice picker's Common Library now includes SONA PRO — a curated collection of premium studio voices, currently spanning Ukrainian, English, Spanish, Japanese, Polish, Russian, and Norwegian. They carry an amber PRO badge, lead the list in their language, and a new PRO toggle next to the Male/Female filter shows only them (with a count for the selected language). Play previews the voice's official sample.

PRO voices work everywhere the Cartesia picker does — the create page, favorites, and the editor's narration voice modal — and narration with them is billed at the normal Cartesia rate. See Narration voices for the full picker walkthrough.

Cards on the dashboard used to show a frame grabbed from the video and the raw prompt you typed — which made a page of projects hard to scan once you had more than a handful.

Now, if a project has generated thumbnails, its card uses one as the cover. The variant you Mark as final wins; otherwise it's the first variant. Hovering the card still previews the video underneath, as before.

And if you generated a YouTube title for the project, the card shows that title instead of the prompt. Your original prompt is still there as the tooltip when you hover the text.

Projects without thumbnails or without a generated title look exactly as they did.

Opening your project list used to quietly start streaming every card's finished video in the background, just to draw the preview thumbnail. On a full page of projects that was several gigabytes of traffic competing for your connection — so pressing Download on a card sat on a spinner for a long time before the browser showed anything.

Cards now load only the little bit of the video they actually need for the preview, and the download button hands the file over the same fast path the project page already used. Downloads from the dashboard now start about as quickly as they do from the project page.

The runtime shown on each card is also now the exact length of the finished file, read once when the video is rendered, instead of being measured in your browser every time the page loads.

Saving your current settings as a preset used to leave the My Presets menu showing the list it had loaded when the page opened — your brand-new preset only appeared after a page reload, and even then it sat at the bottom of All presets rather than near the top.

A preset you just saved now appears in the menu straight away, at the top of Recent — no reload needed. Opening the menu also re-checks the list, so presets saved in another tab show up too.

Bulgarian, Indonesian, Malay, and Tamil have been removed from the output language selector. VidNext synchronizes narration to on-screen text by measuring exactly when each word is spoken, and no speech-alignment model exists for these four languages — so videos in them could not be timed correctly and would fail late in generation, after narration had already been produced. Rather than let a video get that far and break, they are no longer offered.

No existing project is affected: none had ever been created in these languages. The selector now offers 27 output languages, and every one of them supports full word-level narration timing.

Pressing Download on a thumbnail variant — in Thumb Lab or on a project's Thumbnails tab — used to just open the image in a new browser tab, leaving you to right-click and save it yourself.

It now starts a real download straight away, the same as Download video does, and the file is named after the variant's concept (for example pointing-at-window-mould.png).

The Quality setting in the thumbnail preset editor — where a preset could choose which image model generated its variants — has been removed.

Every thumbnail run now uses the same image model, the one the service has configured for thumbnails, and automatically falls back to the next one if it is unavailable. Your existing presets keep all their other settings; only the model choice is gone.

Nothing changes about what you pay: thumbnails have always been charged for the model that actually produced the image, and the credit estimate shown before you press Generate already matches it.

In non-English videos, a word-card motion scene (the letter-series style that reveals one word per card) could wrongly discard its correctly generated on-screen words when the narration used a different grammatical form of the same word — common in Ukrainian, Polish, and other inflected languages — and show built-in English sample words instead. The content check now recognizes inflected word forms, keeps real generated content, and if generation still fails, rebuilds the cards from your own scene plan and script (or falls back to a text card with your narration). Template sample text can no longer appear in a finished video in any language, and affected scenes in existing projects fix themselves the next time the visuals step is re-run.

Card timing in non-Latin languages is fixed too: word cards, letter grids, map stops, and stat reveals now sync to the exact moment their word is spoken in Ukrainian and other non-Latin scripts (previously they could all flash at the start of the scene with the last card frozen for the rest of it). When a card's word is a translation or paraphrase of what the narrator says, an extra AI alignment pass matches it to the right spoken moment.

The pricing page now answers the most common questions right below the packages, grouped into five sections: Product & Quality, Credits & Usage, Billing & Payments, Commercial Rights & Watermarks, and Support & Refunds — covering what the visuals are made from, which formats and languages are supported, using your own script or voice, what a finished minute costs in credits, and how to reach support.

The short version of what it covers: credits are spent on AI generation (script, voiceover, visuals) — timeline edits, caption tweaks, and footage swaps are free, with only the export of an updated video using a small amount of credits for rendering; credits bought during the beta never expire; while the beta lasts VidNext runs on flexible top-ups, with monthly subscriptions arriving after it ends; you keep full commercial rights to your videos, and exports carry no watermark. The section is available in both English and Ukrainian.

Some videos generated since August 10 could open in the editor with a few scenes missing from the timeline — most visibly the very first scene — even though the finished video itself was complete and correct. This affected scenes whose footage was sourced from the stock library during generation.

The gap is fixed, and existing projects heal automatically: the next time you open an affected project in the editor, the missing scene blocks are restored in their exact positions with their original footage. Your edits are preserved — the repair only fills the empty gaps, and it never re-adds a scene you deliberately deleted.

Motion-design templates ship with English demo content so they look complete in the design library — and in rare cases a fragment of that demo text could end up in a finished video when the AI intentionally left an optional field empty (for example, a stat card about a topic with no year-over-year comparison could still show a sample "12.5% vs last year" badge). Optional decorative fields now stay empty when the plan leaves them empty: the badge, attribution line, or label simply doesn't render. Demo content is used for library previews only. This applies to every output language.

Videos generated in Japanese, Chinese, or Korean previously failed during script generation — the length check counted these scripts as a handful of words because they don't separate words with spaces, so every attempt was rejected as "too short". Length measurement is now script-aware end to end: scripts in these languages generate at the correct length for the requested duration, scene pacing is sized in characters instead of words, and the narration voice's speaking pace is calibrated on text in the video's actual language.

Captions got the same treatment: Japanese, Chinese, and Korean subtitles now render as natural phrase chunks without spaces between characters, break at the right punctuation, keep to comfortable line lengths, and use proper CJK fonts (including correct Japanese glyph forms instead of Chinese variants). Titles and text animations size their layouts correctly for full-width characters.

Voice previews now speak your selected content language. When you pick a language on the create page, the voice + music sample and the voice auditions in the picker are narrated in that language for all 32 supported languages — no more English preview for a Japanese voice. The English preview script also got a small refresh.

The VidNext blog has moved to the main domain: it's now at vidnext.ai/blog, and the Blog links across the site point there. Existing links to blog posts on the docs domain keep working — they show the same content — but the main-domain address is the one to share and bookmark.

Music video mode has no voiceover — the uploaded song is the whole soundtrack, and the prompt only directs the visuals. Until now, pasting a full narration script there produced a silent beat-cut montage: technically a valid music video, but not what the script's author wanted.

VidNext now checks for this before the video starts. If you're in Music video mode and your text reads like a script meant to be spoken — whether it's pasted into the prompt or attached as a custom script — a dialog explains what the mode will actually do and shows the first line that would never be voiced, alongside your uploaded track.

From there, one click on "Make a narrated video instead" converts the setup for you: your text becomes the narration script, your uploaded track is kept as background music, and the form guides you to pick a voice before generating. Prefer the montage? "Keep Music video mode" continues exactly as configured — the warning never blocks you.

A live hint under the prompt field also appears the moment you paste long text in Music video mode, so you don't have to reach the Create button to find out.

Videos narrated with Cartesia voices now generate their voiceover per act instead of scene by scene. Narration flows smoothly across scene changes with no clipped joins, and every word gets a precise timestamp — so captions, karaoke-style highlights, and word-synced text animations line up with the voice exactly, the same way they already do for ElevenLabs and Fish Audio voices. Burned-in subtitles on Shorts now work with Cartesia voices too.

Redoing the narration step is also cheaper for these projects: acts that didn't change are reused instead of being generated (and billed) again.

In the editor, the narration voice picker now has a Cartesia tab — browse the catalog (pre-filtered to your project's language), preview voices, use your cloned voices, and regenerate any narration with them. And regenerated narrations now carry their own word timings, so "Generate from narrations" in the captions editor stays in sync with the new audio instead of the original take.

Voiceovers no longer stumble over dotted abbreviations. Text like "U.S.", "F.B.I.", or "e.g." used to be read with a long pause after every period — the voice treated each dot as the end of a sentence. Narration is now prepared for speech the same way numbers already are: the voice reads the letters smoothly ("U.S." as "U S", "F.B.I." as "F B I") or expands small abbreviations the way a narrator would ("e.g." as "for example"), in your video's language — while captions, typewriter, and all on-screen text keep the original spelling exactly as written.

Also fixed in the same pass: money amounts with a magnitude word — like "$220.2 million" — could occasionally be voiced with the magnitude doubled ("…million dollars million"). Amounts now read exactly once, the way a narrator would say them.

This applies to every voice and every video type — landscape videos, Shorts, and narrations regenerated in the editor.

On the dashboard, a project's status now sits in the top-left corner of its thumbnail — the first thing you see — and the "how long ago" timestamp moves to the top-right. Projects paused for script approval, or still waiting in the queue, now say so on the card instead of showing a raw state name.

Also: the pricing page drops its film-footage backdrop for the same drifting studio lighting as the home page, in both the light and dark themes.

The studio backdrop's ambient light has been recomposed: cool brand-blue now anchors the top-left corner behind the navigation, while the warm tungsten glow settles into the bottom-right. Same film-style atmosphere, better balance across the page.

Also: when a video is assigned to a Planner channel, the project page's status bar now shows the channel name — click it to jump straight to that channel's dashboard.

The VidNext website (homepage, pricing, about and careers pages) now offers a light theme. A theme switcher in the navigation lets you pick System, Light, or Dark — by default the site follows your device's appearance setting, and your choice is remembered across visits.

The light theme is a warm paper look built around the same film-set identity as the dark one. Product demos, example videos and screenshots keep their true dark interface — the VidNext app itself keeps its cinematic dark look.

VidNext now sits on a warmer, film-style backdrop: the page background has a warm graphite tone with soft drifting light — warm tungsten pools mixed with cool blue ones — behind the dashboard and creation pages. The marketing homepage and the docs site share the same warmer look, and the homepage's animated product demos now show the interface's new colors.

The dashboard header is more compact too: the four project counters (total, completed, in progress, failed) now live as small chips right under the welcome title, and they still work as one-click project filters.

Also: the Ukrainian community Telegram link on the homepage and in the footer now appears for visitors from Ukraine, and the footer now links to VidNext on X.

The official VidNext documentation is now live at docs.vidnext.ai — guides, billing details, changelog, and blog in one place.

Clip selection got noticeably better at staying on-subject. Candidate footage is now scored more strictly against your video's actual topic, so scenes are less likely to pick up look-alike clips from unrelated stories — especially for videos about specific people, events, or franchises.

The YouTube metadata generator now researches competing videos before writing your titles — so title options, tags, and SEO keywords are grounded in what's actually ranking for your topic, not just the script.

Also in this update:

• Ukraine pricing: purchases from Ukraine now get an automatic discount on every top-up, not just the first — applied automatically at checkout.
• Narration fixes: numbers and currency amounts are now read out naturally by narration voices.
• Language selector: fixed cases where the interface and output language selectors disagreed.
• Editor stability fixes.