v3.0.0 31 Oct 2026 Latest

New tool: Doodle Maker. It has its own place in the sidebar, called Doodle, its own home, wizard and timeline, and a hand draws your whole video on a board as it is spoken: pictures sketched and coloured in, words written letter by letter, icons, arrows and charts drawn as they are said, and a camera that moves on to each new thing and pulls back to show the board at the end.
Start from a script for a voice to read, or from your own recording. Pick the board, the hand and how pictures are coloured with a live preview, then choose the shape, the narrator and the line-art style. Pictures are made as AI line art, and any of them can be replaced with your own on the Boards step.
A doodle video goes through the same steps as any video: Voice, Plan, Boards and Timeline. Generate and preview the voice first, so the new Doodle Director plans on when each word is really said. It reads your whole script once for what it is about, what its words mean here, its one idea and a visual thread, plans each board knowing the boards around it, so a container in a shipping video is drawn as a shipping container, then composes every board like a designed page: up to three pictures side by side with a short label under each, titles, lists, and bar charts drawn when figures are compared. It then reviews the video: no picture is drawn twice, figures it missed are written, figures given year by year become a chart, and no long stretch goes by with nothing new. Words are written strictly in reading order. A video needs far fewer pictures than an ordinary one.
The Boards step shows every board before any picture is made. Change what is drawn and on which word, add or remove things, swap in your own picture, or split and merge boards (which makes their voice again). No picture is made and no credit is used until you press Auto fill.
Plan again on the Boards step asks the Director to plan just that board, with a fresh take that keeps to the rest of the video. Every other board stays exactly as it is, and if the Director cannot answer, the board keeps its plan. A board that already has pictures asks first, since its pictures will need making again.
Every board on the Boards step plays in its own preview with its voice. Make the whole video with Auto fill, or go board by board with Generate assets and Draw board. A live count shows how many pictures are made, then how many boards are drawn. Auto fill can be stopped and continued. A picture that fails is written as words until you try again.
The Doodle Maker timeline has only what a doodle video needs: boards, what is drawn, captions, voice, music and the drawing sound. Drag anything on the Drawn track and it lands on the nearest word. Captions have every option, in the wizard and here: font, size, colours, highlight and its background, animation and position. Captions, music, the drawing sound and the board itself are set in the panel beside the picture, and the boards lay themselves out clear of the captions.
Eleven boards: whiteboard, paper, ruled notebook, kraft paper, chalkboard, glassboard, blueprint, legal pad, graph paper, dot grid and parchment. Four cameras: one thing at a time, close-ups, whole page and a continuous pan. A hand in three skin tones with a choice of sleeve, the pen on its own, or your own uploaded hand. Boards can move on or be wiped by a felt eraser, and the drawing order of any picture can be changed.
The drawing sound is made from the hand's own movement: a marker's rasp, a pencil's hiss, chalk's grit. It sits well under the voice and dips whenever someone speaks. Set it to Off, Subtle or Normal.
A picture and its note can now share one screen: the picture on top with the whole note written under it, however long the note is. The Doodle Director chooses this where the words explain the drawing, and Screen on the Boards step sets it for anything yourself.
Doodle videos can now use more pictures, up to about 6 in every 10 things drawn, depending on the script.
Charts in Urdu and Arabic doodle videos now run from the right.
Your own or stock video on a board is drawn from a still frame, then plays in place.
Almost 1,000 icons to draw, searchable by name on the Boards step, and the Director picks from the ones your words name.
A new Doodle group in the picture styles holds every style a board draws well: Whiteboard Drawing, Stick Animation and the stick-figure, doodle, chalk and calm-science line-art styles, plus eight new ones: Marker Cartoon People, Doodle Icons, Colour Marker Illustration, Continuous Line Art, Pencil Line Art, Technical Line Drawing, Ink Engraving and Crayon Doodle. Each draws one subject on plain white, so the hand draws it cleanly on any board, and they work in ordinary videos too. A doodle video lists them first.
Doodle videos work in Urdu and Arabic. The hand writes from the right, line by line, in Nastaliq for Urdu and Naskh for Arabic, and goes back to dot each word as a writer does. The Doodle Director writes its notes in the language of your narration and describes its pictures in English for the image model. Other languages keep their own words too.
Doodle videos have their own rules for written words. The Doodle Director writes notes as a teacher writes on a board: short labels of one to five words in clear words of its own, up to about 12 for a big line, and each makes sense alone. Normal videos still only show words the narration says. A note stays with what its board says and never adds a figure that is not said, and a note that would run too long is left off whole, never cut. A long note gets a page to itself, wraps to at most three lines on a wide board, and the next thing waits until it is written. Your own words on a board are yours to write freely.
Dots on letters, as on an i, are now drawn by the hand instead of appearing at the end.
The Timeline screen has a new look. The viewer, playback bar, tracks and settings panel now sit together as one editing area that follows your Light or Dark theme, and tracks are grouped into Picture, Cuts and Sound. Voice, music and your own audio clips show their waveforms, and the top bar keeps preview controls on the left and export on the right.
One playback bar now works for both Rendered and Live previews, with frame-by-frame steps, a volume control that only affects the preview, and Space to play or pause. Drag the line above the tracks to give them more room, and use the zoom slider or Fit to see the whole video. Sections in the settings panel fold away with a click and stay folded.
New photo treatment: Sketch reveal. A hand draws any picture on a board, line by line, then its colour comes in and the picture settles under the usual slow push. It works on every kind of picture: line art and stick figures are drawn along their own lines, flat cartoons edge by edge, and photos, 3D renders, paintings and dark night scenes are sketched first, with shading in their darkest parts, before the real picture washes in under the lines. The clip always ends on your picture exactly as it is.
Pick it on the Visuals screen, under photo treatments, and choose how it is drawn: the pen (black marker on a whiteboard, pencil on warm paper, or chalk on a slate, where the colours turn soft and dusty), how the colour comes in (washed in under a brush edge, faded in, or Sketch only to keep the drawing), and what draws (the pen on its own, a hand holding it in a light, medium or dark skin tone, or nothing, so the lines draw themselves). Use on all pictures carries those choices to every picture.
Sketch reveal costs no new image: it draws the picture the shot already has. The drawing is worked out in the background, so the app stays responsive while a picture is prepared, and the live preview draws it exactly as the export does.
The AI Director only uses Sketch reveal when you tick it under Treatments it may use, since it changes how the picture itself looks. With nothing ticked it keeps taking turns with the other treatments as before.
The Visuals screen now shows drawing as well as pictures. A Clips drawn meter beside Shots ready counts every text card, chart, map, treatment, heading, hook and doodle board, and names the one being drawn. Each scene says Drawing, To draw, Drawn or Could not draw, and the Coverage bar marks what is still to draw.
Editing a shot no longer stretches its card across the screen. Photo treatment, text card style, chart and map editing, and stock search now open in a dialog, with the shot and its words on the left and the scene's other shots one click away.
New: subtitle files for every kind of video. Exporting an AI video, a Doodle video or a SceneSync video now also saves an .srt beside the MP4, with the same name. The new Subtitles button beside Export saves one without rendering anything. Subtitles follow the usual rules for subtitle files, whatever your captions look like: whole sentences where they fit, two lines at most and no more than six seconds on screen. They hold everything that is said, even where captions step aside for a graphic or are switched off. Urdu and Arabic are marked to read right to left. A line with a voice but no word timings is spread over its scene by length. A video with no voice has no subtitles.
The .srt written with a CapCut hand-off is made the same way.
Fixed: a WAV recording from some voice generators had its voice cut from the wrong place, so each scene spoke a later line than its captions and the last scenes had no voice at all. These files state the wrong audio speed in their header; the app now corrects that when you bring the recording in, without changing the sound. Make the project again from the recording to get the right voice.
A scene that comes out of a recording with no sound now stops with a message, instead of leaving a silent scene.
Fixed: a recording with little punctuation, as Urdu often comes back from transcription, no longer becomes one scene minutes long. A stretch too long for one scene is now cut where the speaker pauses. Make the project again from the recording to get the new cut.
Fixed: a normal video could open as a Doodle video after restarting the app. Videos it happened to go back to being normal videos by themselves when opened.
Fixed: planning no longer starts a video over when a few scenes were missed. The Plan screen now names the scenes the plan skipped and the shots it wrote no prompt or search words for, and Plan missing scenes does only those, so every other scene keeps its shots and pictures. Each scene on the Plan screen, and each scene card on the Visuals screen, can also be planned again on its own. Plan all again still plans everything, and asks first when pictures would be lost.
Auto mode now asks again for any scene or prompt the plan missed before it fills that scene's pictures, and says on the bar if some are still missing.
Fixed: long videos no longer sit on Generating with no pictures when character portraits fail. Auto-fill now asks first when characters in your shots have no portrait: generate them with progress and any errors shown, or choose Use style only, which fills the pictures from your style and each character's written look and is remembered for that project (Ask me again is in Cast & style). A failed portrait in an unattended run is tried once instead of before every picture, and Stop stops it.
Fixed: character portraits, the style lock and the Consistent look switch are now saved as soon as they change, so a restart no longer loses portraits already paid for.
Fixed: an export no longer fails near the end because of a missing file. Every picture, video, voice, music file and layer is now checked before rendering starts, before you choose where to save and before a credit is used. If any file is missing, the export stops straight away and lists each one with its scene and shot. Text cards, opening hooks, photo treatments and doodle boards whose files were lost are drawn again by themselves. A shot with no picture, a scene with no voice or missing sound effects will not stop an export: they are listed first, and you can fix them or choose Export anyway. Scene previews make the same check.
The Dashboard is redesigned around all your work. AI videos, Doodle Maker and SceneSync each have a way to start from it, and they share one view: what to continue, what is up next, totals across every tool, a chart of projects started by tool, and recent projects you can filter by tool.
The Export window now shows whether your video is rendering on the GPU or the CPU, which graphics card and encoder it is using, and the export size. If the GPU fails partway and the export moves to the CPU, the window says so. When the export finishes, the video plays right there in the window, and Open folder takes you to the file. If the app is in the background when it finishes, the video starts muted.

v2.7.0 26 Oct 2026

New: templates. A new Template step in the new-project wizard asks what kind of video you're making: Documentary, History, Mystery, Top 10, Shorts explainer or News explainer. One pick sets up the whole edit the way that kind of video is cut: the frame, the cut rhythm, the opening hook styles, chapter titles, time jumps and re-hooks, which graphics and photo treatments the Director may use, the colour look, the map style, the headline type, whether captions step aside for graphics, and the sound on the cuts. Your own settings is still there for setting it all up yourself.
Under the templates, What the Director will choose lists every one of those choices with a switch, so you can see what a template does and turn any part off before the project is made. Reset puts a template's choices back after you've changed them. Which picture sources the Director may use (AI images, stock photos, stock video) is never changed by a template, since that decides what a video costs.
The Review step shows the template you started from, and what you've changed since.
Each template now tells the AI Director what kind of video it is making, not only which settings to use: a History video shows the period's clothes, buildings and streets, a Mystery builds tension without giving the answer away, a News explainer keeps to real places and a chart for every figure, and so on. With no template picked, a script that clearly reads like one gets that template's guidance anyway.
The Template step reads your script and suggests the template it reads like: "Your script reads like a Top 10: 10 numbered entries found, counting down from 10". Use it with one click, or ignore it.
Top 10 now works for ranked lists of any length (a Top 5, 7, 15 or 50) and however the entries are numbered: "Number 10 — Chernobyl", "No. 7: Pompeii", "#3", "10. Venice" on its own line, "At number seven, we have…", "In fifth place…", or "And finally…" for the last one. It counts up as well as down, and works from a recording too, including transcripts that spell the numbers out or have no punctuation.
In a ranked list, every entry starts a scene of its own, and its number and name go up big over its first picture: the entry's own number, not a chapter count. Before, a real Top 10 showed "02" while the voice said "Number 9", and four of its ten entries had no title at all. An entry with no picture to carry its title (all video, say) gets a rank card instead, so no entry goes unnumbered.
A heading line like "Number 9 — The Great Hanshin Earthquake, 1995" now gets a full stop, so the voice pauses after it instead of running it into the next sentence.
The opening hook of a ranked list counts up to the list's own number: "Top 10 · Deadliest Disasters in History". Before, it counted up to the year of entry number 10.
Cards the Director makes to announce an entry ("#6 …", a "Number 4" stat, "2 — Yellow River Flood") become a picture of that entry with its title on it, so every entry looks the same. A number that names something ("Number 1", "Top 10") is no longer counted up as a Big number or charted.
Re-hooks no longer go missing when the half-way point has no hook line or question: the strongest line said there is brought back instead.
New story mark: name tags. Each person the narration introduces with a role, such as "President Kennedy", "Elon Musk, the CEO of Tesla" or "the captain, Benjamin Briggs", gets a name tag with their name and role, as said, on the first picture of them. Only a picture that is of them gets one: never an empty room, their desk, or a newspaper about them. On for the News explainer, Documentary, History and Mystery templates, and a switch under Story marks for any video.
Mystery videos keep their answer for the moment it's revealed. The Director is told where the narration reveals what happened ("Then, in 2006, researchers finally concluded…") and shows nothing of it before then. Re-hooks bring back the questions the narration asks, and only before the answer, and the opening hook never quotes the answer.
History videos get a timeline where a line says two or more dates, at most one scene in three, as long as timelines are allowed.
New in the Script step: lines copied from an AI chat or a web page, such as a source's name over a "+1" chip, footnote marks like [3] and bare links, are pointed out before the voice reads them aloud ("University of Maryland School of Medicine plus one" was voiced four times in a real Top 10). Remove them takes them out in one click.
The new-project wizard now also remembers the template you used, your headline type and your colour look, so a project made after a Top 10 starts as a whole Top 10 again, not half of one.
New in the AI Director: story marks, headings the Director puts where the story turns. Each is a heading on a picture, in your type and accent, with words the narration says, and plays over the shot while the narration carries on.
Chapter titles: a big number and the chapter's title where each chapter of the Story Bible starts. The title is the Story Bible's own when the scene says it as a whole phrase. A title too long for the title goes whole under the number, and a title that stops mid-phrase is never used: a whole phrase the scene says takes its place. Chapter titles are at least 20 seconds apart, so a short gets one or two, not one per scene. They need Story continuity, which finds the chapters.
Time jumps: a place stamp where the narration jumps in time, such as "Three days later", "The next morning" or "On October 22, 1962", over the picture it is said under. Only the start of a sentence counts, so a date in the middle of a line stays a fact, not a jump.
Re-hooks: a strong line brought back on screen to hold the viewer, about every minute, 75 or 90 seconds, or once half way for a short. It is a line the Story Bible marked as a hook, or else a question the narration asks.
Switch them on in the Director settings, under Story marks, or let a template do it. Story marks never go on two scenes in a row, never on a video or a graphic, and never over your own photo treatments or headings.
The AI Director now balances its own edit after planning. A heading right beside a story mark, or the same photo treatment twice running, is taken off. Three graphics in a row are pointed out in the planning notes, since which one should be a picture is your call. Only the Director's own picks are changed, never yours.
New on the Visuals screen: a Story marks panel above the scenes shows how many chapter titles, time jumps and re-hooks the video has. Edit switches each kind on or off and changes the re-hook rhythm. Place story marks now places them all again over the shots as they are, with no AI call and no new pictures, and Take all off removes them while keeping your own treatments.
New on the Visuals screen: each chapter of the Story Bible starts with a divider showing its number, title and scenes. Re-plan this chapter asks the Director for new shots in that chapter's scenes only, and asks before replacing anything.
A treatment the Director placed as a story mark says so in its panel, and edits like any other heading.
Suggest's phrases from the narration no longer break before "to" ("was still | to come"), and a whole sentence such as "Nobody is coming to save you" can now be suggested.
Fixed: a long voiceover could be cut into scenes that didn't match the recording. When the transcript and the word timings disagreed, as when a recording's dollar amounts were missing from the transcript, the scenes after that point slid later and later. On an 8-minute recording a five-word scene got 31 seconds of audio, and the last 500 words were squeezed into the final seconds. Scenes now stay on the words as you say them, however much the two disagree. Projects already made keep their scenes: make the project again from the recording to get the new cut.
Fixed: transcripts losing figures. Amounts and percentages the speech service heard but left out of its text, such as "$3,000" or "50%", are put back, so captions and the AI Director see what you said.
Text on screen now waits for the words. A heading, chapter title, time jump or re-hook only goes on a picture where the words are said within two seconds of it appearing, and only if they can be read before the cut, at about 200 words a minute. A re-hook too long for its picture is cut to a whole phrase of it that fits, and a heading that can't be read in time is left off. A chapter title the Story Bible can't give whole shows its number alone, never a stray phrase from later in the scene.
Text on picture warns you when a heading has more words than its picture is on screen long enough to read.
A chart now starts with the clause that says its figure, not a sentence before it, and a chart that ends its scene comes up at most a second early instead of a whole five. A chart that counts up a single number no longer takes it from inside a name like "the 50-30-20 rule".
Text cards only say what the scene says. A card with a figure the scene never says, or a label the AI Director made up ("20% Allocation"), is no longer put on screen.
Fixed: in the installed app, photo treatments, headings on pictures and the opening hook could not load their picture, so they played as the plain picture ("could not be drawn"). Cards, headings and caption previews also fell back to a default font instead of the bundled ones, and the SceneSync waveform stayed empty. All of them now draw the same as during development. Use Redraw all graphics on the Visuals screen once to redraw any that played plain.
Fixed on Mac: scene cards on the Visuals screen showed a "?" instead of the picture, though the picture opened fine full size. The small previews are kept in a hidden folder the Mac app wasn't allowed to read. It now can, and a preview that still won't load shows the picture itself.
Suggest for text on picture now asks the AI. It reads what is said under the picture along with the scenes just before and after, so it knows what the story is about, and offers up to three headings, each with a label above and a line under where the style has them. Press Suggest another to step through them; when they run out, it asks for new ones. Every word still has to be said in the scene's narration, and words that aren't are left off.
Fixed: Suggest often gave no words at all, because it only took a clause short enough to fit whole and most sentences are longer. Without the AI (no key, or offline) it now finds the whole phrases inside a long sentence, such as "the town of Pripyat was abandoned", still never cut off mid-phrase.
A Chapter title fills in its number by itself, counting the chapters before it, and Use on all pictures numbers them in order.
Suggest only replaces a label or line under that is empty or that it filled in itself. Words you typed stay.
Custom styles now come with a template. Blank template fills the box with one line for each part of a look (Medium, Colours, Lighting, Texture, Mood, Framing), and you replace each [ … ] with your own words. Three finished examples written the same way (Vintage Film, Storybook Watercolor, Neon Noir) can be loaded to start from, and How to write a style explains each part. A style with a [ … ] still unfilled can't be saved or previewed, and loading a starter over words you've written asks first.
Custom styles can now be up to 2,500 characters instead of 600, room for the full template plus parts of your own such as Character Style, Environment or Rendering Quality. A style written in a chat can be pasted in as it is: **bold** part names, bullets and blank lines are tidied away.
Custom styles draw closer to what you wrote. The start of a Medium line now opens every image prompt, the way the built-in styles do ("watercolor painting of …"), so models that read mostly the start of a prompt still get the right kind of picture. The template and examples now ask for light that changes with the scene, so a style no longer puts every shot at the same time of day. The guide explains why a style should say how things are drawn rather than what is in the picture, and that quality words like "4K" or "masterpiece" add little; a style that uses them gets a gentle note.
A new export window. The bar now moves steadily from start to finish. Before, it could sit at 0% while the export got ready, run to 100% on the scenes and then fall back to 50%, and stay at 70% for the whole caption burn, which on an hour-long video looked like a hang. Every step now fills its own part of one bar: each scene as it renders, and captions, layers and sound by how far through the video they are ("41:10 of 1:02:00"). The window shows the elapsed time, an estimate of the time left, and steps for Prepare, Scenes, Join, Captions and Sound. Progress also shows on the taskbar, so you can minimize the app. When the export finishes, the app flashes if you were in another window. Cancel asks before stopping, and the finished screen shows the file, its folder, how long the video is and how long it took to render.
Fixed: Auto-fill could sit on a scene for minutes, which looked like the app had hung. A stock clip from a slow archive could hold up the fill for up to 15 minutes before it failed. Now a clip that stops arriving for 30 seconds, crawls along, or whose server doesn't answer is swapped for the next best result, usually from another source, so the shot still gets a picture. A shot that still has nothing after 3 minutes (90 seconds for a photo) is skipped and listed so you can fill it again, and the rest of the fill carries on.
Auto-fill no longer waits on the slowest stock source. A search goes on with the sources that answered within 15 seconds.
In AI Director projects that mix footage, AI images and graphics no longer wait in line behind stock videos. Each kind fills at its own pace, and the limits now hold across the whole run, so a long script doesn't send more downloads at once than your connection or the providers allow.
Auto on a single scene fills its shots side by side instead of one after another, and one bad shot no longer stops the rest.
Stop now stops right away. Downloads and images in progress are dropped at once and aren't listed as failures.
The run bar says what it's waiting for, so a pause never looks like a hang: "waiting for scene 3's narration", "planning shots for scenes 5–8", "a clip was slow, trying another", "drawing the text cards".

v2.6.0 19 Oct 2026

Better-looking text cards. Cards now have their own headline type instead of borrowing the caption font: pick Impact, Clean, Condensed, Heavy, Editorial or Typewriter in the Motion graphics panel, or keep Caption font. Each pairs a headline face with a plain face for smaller text.
Headlines are set more like a designer would set them: capitals get a little letter spacing, lines sit closer together, and the words have a soft shadow so they lift off the background.
Each line of a card now rises into place from behind its own edge, one after another, and the card's words leave again just before the cut instead of being cut off. Very short cards, and cards held longer than 12 seconds, keep their words to the end as before.
A card can carry a small label above its text ("The truth", "Part two") and a line under it. Add them on the Visuals screen with Label above and Line under, or let the AI Director write them. Any number in them must be one the narration says, or it is left off.
The default card background is now a near-black with a hint of your accent colour. Before, it was your accent darkened, which turned yellow into olive and red into maroon. A background you picked yourself is unchanged.
Text the headline faces can't show, like Urdu or Arabic, is set in your caption font, as before.
Every existing card is redrawn once in the new style. Cards with a long hold take a little longer to draw the first time, because the clip now runs to the cut.
New: text on picture. Put a heading straight onto a scene's own picture: a Headline low on the frame with a slow push-in, a Chapter title with a big number over a greyed picture, a Place stamp typed in with a pin, a Name tag that slides in, a Quote over the darkened picture, a Split screen with a headline panel, a Big number that counts up, or a Boxed headline with each line on its own white box.
Find them on the Visuals screen, under Text on picture in the same panel as photo treatments. Most take a small label above and a line under, as well as the heading.
Every word in a heading must be said in the scene's narration, just like photo treatment captions. Suggest picks words from what is said under the shot, and words the narration doesn't say get a warning and are left off. A chapter's number is the only exception.
Headings use the same type as your cards and your accent colour, never make new images, and the live preview draws them exactly as the export does. Each has a light sound of its own.
New in the AI Director: Headings on pictures, an opt-in switch below Photo treatments. The director puts a heading on at most one picture in five, never two close together and never on the picture right after a photo treatment. A figure that is said becomes a Big number, a question becomes a Quote, a date becomes a Place stamp, and otherwise Headline, Boxed headline and Split screen take turns. Name tags and chapter titles are only added by you.
A date is never shown as a Big number ("April 28" no longer counts up to 28), and a heading is always a whole phrase from the narration instead of being cut off mid-sentence.
New: captions can step aside for text graphics. A new Over graphics setting in the caption options: Show keeps captions everywhere, Skip text hides them while a text card or a heading on a picture is on screen, and Skip all also hides them over charts, maps, timelines and photo treatments with words on them. The narration still plays; only the captions are left out, so they never sit on top of other words.
A caption is split where it is skipped instead of carrying on across the card, and the live preview hides captions exactly where the export does.
New projects start with Skip text. Existing projects keep showing captions everywhere until you change it.
Bundled a new font, DM Serif Display, for the Editorial type.
Fonts that come in one file with every weight, such as Montserrat and Inter, are now drawn at their true weights in cards and headings, instead of a faked bold.
New in the AI Director: Story continuity, on by default. Before planning a script, the Director reads the whole of it once and writes a Story Bible: what the story is, its look, the places it happens in, and how it divides into stretches of place, time and mood. Every group of scenes is then planned knowing the whole story, the shots just before it and the lines coming next, so a place looks the same every time it appears and the mood follows the story instead of changing between groups.
Each place in the Story Bible has a fixed description that is added to every picture set there. Open the Story Bible on the Plan screen to read it, change a place's look, the logline or the palette, or read the script again. Your edits are kept until the narration itself changes.
With consistent characters on, the Story Bible also finds the people in the story and adds them to your cast with a description of how they look. On Runware and the Vidoptimus Image Server, each of them gets a reference portrait first, which then goes with every shot they are in so their face stays the same. With consistent characters off, nothing is added and prompts work as before.
Consistent characters can now be turned on in the new-project wizard too, in a new Characters panel on the Models step. Turning it on there or on the Visuals screen moves a provider that can't read photos to one that can.
New in the Cast panel: Find characters reads your script and adds the people in it, each with a description of how they look. It works on scripts of any length; with consistent characters on, planning the shots also finds them.
New in the Cast panel: Generate draws a portrait of a character in your chosen style, from their description, with the image model picked on the Visuals screen. With Style lock on, it also matches your style picture. The portrait becomes their main photo, and Regenerate makes a new one. Generate all does every character who has no photo yet.
The Story Bible also marks chapters and the strongest hook lines, always quoted from your narration.
The 26 Niche styles each have their own fixed look, so the Story Bible keeps to the style you picked instead of inventing one.
After planning, pictures that come out nearly the same as one already in the video are planned again once, with something different.
Scenes the Director leaves out are now asked for once more before they fall back to a single picture.
The AI Director now plans like an editor who has read the whole script. After the Story Bible, one more short read marks where the video should cut from pictures to a graphic: a chart where a figure is said, a map where places are named or a journey is made, a timeline where dates are said, and a text card on the lines that land hardest. Each scene is handed to the Director with that note, and every note is checked against the scene's own words first, so a chart only shows a figure that is said, a map only places that are named, and a card is always a whole phrase. Cards stay rare (at most one in three scenes), a figure said in the narration always gets its chart, and a short last line closes the video on a card. Story Bibles written before this update are read once more to get the plan.
AI images in a story set in the past now name its period, so a 1986 street gets 1980s cars and clothes. The Director no longer draws maps, charts or written signs as AI images, and words it asks to be written inside a picture are taken out, since AI images garble them.
Every script gets a Story Bible while Story continuity is on, short ones too. The Story Bible only shapes how the pictures look: charts, maps, timelines and text cards are still used wherever a line has figures, places or dates, and the Story Bible's hook lines are the first picks for a text card. If the Story Bible can't be written, planning carries on the old way and says so. Turn Story continuity off in the Director settings to plan each group of scenes on its own, as before.
Requests are now sized to what your API key allows. A paid Groq or OpenAI key plans up to ten scenes per request and three parts of the story at once; a free key keeps requests small enough to be accepted.
New: the API keys screen shows what your Groq or OpenAI key allows, read from the provider itself, for example "Groq · Free plan · 8k tokens/min". It is checked when you save a key, or with Check plan.
On a free Groq key, the Plan screen warns before a script that would go past the free plan's 200k tokens a day, and offers to plan it without Story continuity.
A request too big for a free Groq plan now says so plainly and is sent in smaller pieces, instead of showing the raw error.
The Plan screen and the auto run show how much of each run the provider served from its cache, which costs less and on Groq doesn't count against your limits.
The built-in Voicely Voices are fixed.
Fixed: Voicely voices that sometimes sent no audio, such as the default Andrew Multilingual, failing the scene. The take is now spoken again without word timings, and the captions are timed by Whisper as usual.
OpenAI: GPT-6 Luna is now the default model. GPT-4o and GPT-4.1 mini were removed, since they cost several times more for no better plan. Projects that used them move to GPT-6 Luna.
Transcription is now a separate choice from the model that writes your prompts. You can write with OpenAI and time captions with Groq's Whisper, or the other way round. Pick it on the Models step of the new-project wizard, on the Script screen, on the Voice screen, or before you choose a recording. Existing projects keep the service they used before.
New: choose the language your recording is spoken in before it is transcribed. When you bring your own voiceover, a Spoken language list sits under Transcribe with, in the new-project wizard and on the Script screen. It has every language Whisper supports, from Afrikaans to Yoruba, including Urdu, Hindi and Arabic. Auto-detect is still the default; naming the language stops Whisper guessing wrong from a short or accented opening. New projects remember your choice. Hawaiian and Cantonese may only work with Groq. Choosing a recording no longer starts the transcription straight away: choose the file, set the service and language, then press Transcribe. Replace works the same way, with Cancel to keep the transcript you have.
New: Preview voice, on the Voice screen, the Script screen and the Audio step of the new-project wizard. It reads the start of your script in the chosen voice, with your rate, pitch and tuning settings, so you can hear a voice before you voice every scene. Playing it again with nothing changed doesn't call the provider again.
MusaTTS now needs MusaTTS 4.5 or newer. Open MusaTTS, switch the API server on under Settings > API, and paste its API key (it starts with musa_) as the Musa key in Settings. Older versions of MusaTTS won't work with the app.
MusaTTS voices: the 46 built-in voices are listed with their language and gender, and the voice blends you saved in MusaTTS's Voice Blending panel now appear in the same list, marked "(blend)". A blend speaks exactly as it does inside MusaTTS, including its pronunciation fixes and how it reads numbers and dates.
New: MusaTTS Voice Clone Pro in 8 languages: English (three versions), French, German, Italian, Portuguese and Spanish. Pick the language where the model used to be. Each language has 26 preset voices, listed after your own custom voices and your Voice Store clips.
Voice Clone Pro shows what it needs next for your language: open MusaTTS, download the language (about 200MB, once, with its 26 preset voices), or load it. When it's ready, All presets fetches any preset voices that are still missing.
New: make your own cloned voice from the Voice screen. Pick a clean 5 to 15 second recording of one speaker and give it a name, and it becomes a custom voice for that language, selected as soon as it's ready. A custom voice can be deleted from the same place.
A Voice Store clip can be saved as a custom voice with Save as custom. Narration in a custom voice is much faster, because the clip isn't processed again for every scene.
Fixed: Voice Clone Pro voices in languages other than English no longer have English pronunciation rules applied to them.
MusaTTS errors now say what to do, for example that the API key needs copying again, that MusaTTS isn't open, or which language needs downloading, instead of showing the raw response.
Projects that already use MusaTTS or a Voice Store clip keep their voice.
Fixed: photo treatments that showed an empty grey frame in the exported video while the picture looked fine on the Visuals screen. A picture that didn't load the first time was drawn as a grey box and kept. Pictures are now loaded more patiently, and a treatment is never kept without its picture: if the picture still won't load, that shot plays as the plain picture and the treatment is tried again later. Every photo treatment is redrawn once to clear out any grey boxes already made.
New: redraw graphics by hand. The app draws text cards, charts, maps, timelines, photo treatments, headings and the opening hook itself, and keeps each drawing until something about it changes. If one ever comes out wrong or empty in the export, you can now draw it again from scratch.
Redraw all graphics, on the Visuals screen next to the scene filters, redraws every one in the video and shows its progress.
The brush button on a scene card redraws that shot's text card, photo treatment or heading, and the opening hook if it is on that shot. Charts, maps and timelines keep their own Redraw button.
The Opening hook panel has its own Redraw button, and the photo treatment panel has Redraw and Redraw all treated pictures.
Each one says afterwards which graphics, if any, still could not be drawn and in which scene. A photo treatment whose picture won't load plays as the plain picture, so replace that picture and redraw.
AI33 is more patient. A line of speech now gets at least five minutes to finish instead of two, and longer lines get more time. A dropped connection or a busy server while a line or picture is being made no longer fails it: the app waits and asks again. If AI33 is busy when you start, the request is sent again a moment later.
When an AI33 line or picture takes too long, the message now names the task and warns that it may still finish and use credits, so you can check your AI33 history before trying again.
Fixed the app sometimes freezing ("Not Responding"). Drawing charts, maps, cards and photo treatments, importing a picture, video or clip, reading saved API keys and looking for CapCut all used to run on the same thread that draws the window, so a slow disk, a big file or a busy antivirus could stop the whole app until they finished. They now run in the background and the window stays responsive.
EdgeTTS voiceover no longer waits forever when Microsoft's service stops answering. After a time limit that grows with the length of the text, the line fails with a message to check your connection and try again.
Big projects are smoother to edit and during auto runs. Autosave used to re-read the whole project on every small change and re-read every project in your library after each save; it now checks once per pause in editing, still saves at least every few seconds during an auto run, and updates only the project that changed.
If the app does hang, the log file now says what was happening: any step that is taking far longer than it should, and any time the window itself stopped responding, along with which screen you were on. Send the log with your report.

v2.5.0 11 Oct 2026

New: photo treatments. Show a picture as a real object instead of filling the frame: a framed print dropped onto a wood desk, a polaroid with a handwritten caption, torn paper, a newspaper clipping, a scrapbook page, a film strip, an old TV, a slide projector, a magnifying glass, an evidence board, a gallery frame, a torn reveal and an instant photo that develops.
Pick a treatment for any picture on the Visuals screen, with the frame button on its card. Use on all pictures shows every picture in the video the same way, and Remove from all takes the treatment off again.
Treatments never make new images. Each one uses the shot's own picture. The scrapbook, film strip and evidence board also use the pictures from the shots around it, and the torn reveal tears away the picture before it.
A polaroid's caption, a newspaper headline, an evidence card and a gallery plate only ever show words the narration says during that scene. Suggest takes them from what is said while the picture is on screen, and words the narration doesn't say get a warning and are left off.
Each treatment has its own sounds, like a print landing on the desk, a camera shutter or a projector clicking on. They replace the cut's usual sound effect, and they follow your effects on/off switch and volume.
The torn reveal is a transition in itself, so the cut into it has no fade, flash or punch.
The colour look grades treated pictures the same way as your other photos.
The live preview draws every treatment exactly as the export does.
New in the AI Director: Photo treatments, an opt-in switch next to the opening hook. The director shows at most one picture in four as a treatment, never two in a row and never the same treatment twice running, and favours a polaroid or a gallery plate when a date is said. You can limit it to the treatments you like.
Projects without photo treatments render exactly as before.
OpenAI's GPT-6 Luna added to the model picker.
Fixed: OpenAI's GPT-5.6 Luna and GPT-6 Luna refused every request, because the app sent them a token limit and a temperature these models don't accept. They now get the settings they take, with their thinking turned down so the answer has room.

v2.4.0 4 Oct 2026

Opening Hook
The opening hook now plays by itself for 4 seconds before scene 1, so viewers give it their full attention.
During the hook you hear only its own riser and hit sound, with no narration and no music.
Scene 1 now starts with its own first picture, and the narration and music begin together.
The hook no longer replaces scene 1's picture or cuts into its time.
The hook always plays at full length, even if scene 1 is short.
The hook shows as its own block at the start of the timeline, and the music starts right after it.
Previewing scene 1 now opens on the hook, exactly like the final export.
A video with a hook is now 4 seconds longer than before; videos without a hook stay the same.
If you already had a hook and added your own overlay or audio clips, those clips stay where they were, so check they still line up with your scenes.
Colour Looks
New: give your whole video one colour style (a "look") with a single click.
Choose from 8 looks: Cinematic, Warm documentary, Cold thriller, Vintage film, Black & white, Punchy (made for shorts), Noir and Faded matte.
A grid shows your own first picture in every look, so you can see the result before choosing.
A Strength slider lets you use just a light touch of the look or the full effect.
Looks only change your photos and video clips; maps, charts, text cards and the opening hook keep their original colours.
Add film grain (fine or heavy) and a vignette (soft or strong) to make the whole video feel like one piece of film.
Each look comes with its own grain and vignette, and you can change them afterwards.
The grain only affects brightness, like real film, so you won't see ugly coloured noise.
Cinema bars turn a 16:9 video into a widescreen 2.39:1 frame, and are never added to vertical videos.
Captions always sit on top of the look, so they stay sharp and easy to read.
You'll find looks in the new-project wizard, on the create screen, and in a new Look tab on the timeline.
The live preview shows the look as you scrub, and the generated preview matches your export exactly.
Projects without a look render exactly as before.
SceneSync (Use Your Own Voiceover and Pictures)
New: SceneSync turns your own voiceover and your own pictures into a finished video, with its own spot in the sidebar.
Choose your audio (mp3, wav, m4a, aac, ogg or flac), then add your pictures and clips as a folder or as individual files.
Name files in order (01.jpg, 02.jpg, 03.mp4) and they line up automatically; 2 comes before 10, and unnumbered files go last.
You're warned if a number is used twice, a number is missing, or a file has no number.
Reorder pictures by dragging them, typing a new position into their number, or nudging them with the arrows.
Your manual order is kept when you add more pictures, and "Sort by number" brings back the file-name order.
Auto Split spreads your pictures across the audio and changes pictures during the speaker's pauses, never in the middle of a word.
A Pacing slider lets you choose between an even split and one that follows the speaker's pauses.
Check your timing before creating the video: play it back and drag any cut on the waveform to move it.
You're warned if a picture would only flash on screen, or if you have too many pictures for the audio.
Captions are made automatically from your audio using your Groq or OpenAI key, and follow your voice word by word.
Without a key, the video is created without captions.
Choose your video shape (16:9 or 9:16), a slow zoom on pictures, and fades between pictures.
Your video then opens on the timeline, where everything else works as usual: caption style, music, hook, colour looks, sound effects, layers, preview and export.
Your original recording plays in one smooth piece, with no clicks or gaps at the cuts.
A SceneSync video costs 1 credit from the same allowance as AI videos, taken the first time you export it; re-exporting is free.
SceneSync: Exact Match (Set Each Picture's Start Time)
New: tell SceneSync exactly when each picture should appear, down to a tenth of a second.
Put the start time in the file name: 02_00-07.5.jpg starts at 0:07.5, and 14_1-02-03.mp4 starts at 1:02:03.
Use "-" instead of ":" in names, because Windows doesn't allow colons in file names.
Or skip renaming and bring a timing file: a cue sheet (.txt or .csv) with one picture per line, listing a time and the file name, number or chapter title.
Audacity labels work too: press Ctrl+B at each picture change while the audio plays, then export the labels.
Subtitle files (.srt or .vtt) also work: picture 1 starts with subtitle 1, picture 2 with subtitle 2, and so on.
A timing file in your pictures folder is found and used automatically.
If a time appears in both a timing file and a file name, the timing file wins.
Pictures without a time are spread evenly between the timed ones around them, so you only need to time the moments that matter.
A table shows each picture's start, how long it's on screen, and where its time came from.
You're warned about two pictures at the same time, a time past the end of the audio, a cue line that matches no picture, and pictures listed out of order (one button fixes the order).
"Snap to pauses" moves each time by up to 0.3 seconds into the nearest pause so no picture changes mid-word; it's off unless you turn it on.
Without a transcription key, the subtitle file's own text is used as captions.
SceneSync: Transcribe & Guide (Audio First, Pictures Later)
New: if you have audio but no pictures yet, SceneSync writes out the words and splits them into scenes for you.
Scenes are split at the end of sentences, so a scene never starts in the middle of a word.
Let SceneSync pick the number of scenes, or ask for an exact number.
You see the average, shortest and longest scene length.
Join a scene with the next one, or split one in two, wherever you like.
Save a scene sheet: a numbered list of every scene with its spoken words and length, as a text file and a spreadsheet, plus an empty images folder ready to fill.
Choose your video shape first, and the sheet tells you what picture size to make.
Name each picture with its scene number (01.jpg, 02.png, 03.mp4) and each scene appears next to its picture, so a wrong match is easy to spot.
Not ready yet? Choose "Create now, add pictures later" and the video waits on the SceneSync home, which shows how many scenes still need a picture.
"Add pictures" places each picture in its scene by number, tells you about any that didn't fit, and can save the scene sheet again.
This feature needs a free Groq key (or an OpenAI key) to transcribe, and captions come with it, timed to every word.
SceneSync: Beat Sync (Cut Pictures to Music)
New: bring a song and your pictures, and every picture change lands on the beat.
It works offline and needs no key.
SceneSync finds the tempo, every beat and the start of every bar by itself, even if the band speeds up or slows down.
It tells you if a track has no steady beat to cut on.
Choose how often pictures change: every beat, every 2 beats, every bar, every 2 bars or every 4 bars.
Cuts start on the first bar so they follow the music's natural phrasing.
"Follow the energy" cuts faster where the music is loud and slower where it's quiet.
Spread your pictures over the whole song, or loop them so just a few pictures can fill an entire track.
If the tempo is read at double or half speed, fix it with ×2 or ÷2, or tap along to set it yourself.
Nudge the beat earlier or later in steps of 10 milliseconds.
A strip shows every beat, with taller marks at the start of each bar, and any cut you drag snaps to the nearest beat.
Each cut lands with a quick zoom punch, and a flash marks big moments like the drop or chorus.
Cuts are hard by default, and sound effects are off so they don't play over your music.
Punch on Cuts
New: "Punch" in the timeline effects (next to light effects) makes each new picture land slightly zoomed in, then settle in a fifth of a second.
It shows in both the live preview and the export.
It works on any project, and it's off unless you turn it on.
Image Provider: Vidoptimus Image Server
New: make images on your own graphics card with FLUX.2, free and offline, in about 3 to 6 seconds each on a modern card.
To set it up, open Settings, go to AI Images, paste the Server URL and API key shown in the server window, then click "Test connection".
Pick Fast, HD or Full HD as the image size.
It also reads reference pictures, so a style picture and your recurring characters keep their look.
Visuals Screen
The Visuals screen has a fresh look: every scene is now a card with a big picture, in a grid that uses the full window width instead of a narrow list of small rows.
Each card shows the scene number, how long its picture is on screen, and whether anything is missing or still being made.
If the AI Director cut a scene into shots, the card shows a strip of them; click one to see it and work on it.
The card shows the words that each shot plays over, so you can tell at a glance if the picture matches what's being said.
Every tool from the old rows is on the card: switch a shot between AI image, stock photo, stock video and text card, upload your own file, generate or regenerate, auto-pick a stock match, or search stock right on the card.
Text cards keep their full style controls, and charts, maps and timelines keep their data editors behind one button so the card stays tidy.
New "Preview all": step through every scene big, one after another, using Next and Back or the arrow keys.
Each scene shows its narration with the words on screen highlighted, plus the lines before and after, and the shot's prompt, which you can now edit right there.
If a picture doesn't fit, fix it in the preview: regenerate it, re-pick or search stock, upload your own, switch its source, or make it the style anchor.
A strip along the bottom shows every scene, with a red dot on any that still need a picture.
A new Coverage bar under the scenes shows your whole video at once, with each shot as wide as its time on screen, so missing pictures are easy to spot; click a scene in it to preview it.
You can filter the screen to show only scenes that still need a picture.
The opening hook now sits in one slim line above the scenes and opens when you click Edit.
"Consistent characters & style" is now a single switch in the toolbar, with your cast and style lock under a "Cast & style" button.
New Project Wizard
Colour looks in the wizard are now shown on a real sample photo, processed exactly like the export, so you can see the true difference before your project has any pictures.
Once your project has pictures, the looks are shown on your own first picture, as before.
The old Look step is now four shorter steps: Frame (video shape and where pictures come from), Visual style, Colour, and Models.
On the new Models step you choose the model that writes your visual prompts, plus the image provider, image model and stock library, even with auto mode off.
The Visuals screen uses the same choices, and you can still change them there.
The Review step lists Frame, Look and Models separately, and each Edit button takes you straight to its step.
Visual Styles
New: 26 visual styles made for specific kinds of channels, in a new "Niche" category in the style picker.
Photoreal: Battlefield Reconstruction, Age of Sail Cinematic, Expedition Documentary and Creator Vlog Photoreal.
Painted: Museum Oil Painting.
Cartoons: Sitcom Satire Cartoon, Big-Head Business Cartoon, Money Math Cartoon, Hardware Explainer Cartoon, History Stick Figures, Storybook History Cartoon, Crude Doodle Cartoon and Midnight Cyan Vector.
Line and sketch: Chalkboard Line Art, Calm Science Line Art, Naturalist Field Guide, Sketchbook Stick Figures, Pencil Wash Metaphor and Editorial Metaphor Cartoon.
Comic and anime: 2D Comic Explainer, Graphic Novel Fable, Storytime Comic Sketch, Crime Comic 2D, Anime Film Slice of Life, TV Anime Fantasy and Cozy Home Illustration.
Each style describes its full look (medium, line, light, colours, framing and mood), so every picture in a video stays consistent.
All existing visual styles were rewritten the same way, and the lighting now changes with the scene (dawn, noon, dusk or night) instead of staying fixed across the whole video.
Each style now names its own colours and shot types, so pictures match the style better and look less alike.
Your projects keep their styles, but new pictures may look a little different from the ones you already have.
Consistent Characters & Style
Your cast's photos now decide who a character is (face, hair, clothes), but the character is drawn in your video's style, so a photo won't turn into a photoreal face in a cartoon or painting.
A style picture now gives only the look, never its people, faces or objects.
Other people in a scene no longer get your character's face.
Fixed: Consistent characters & style now works with the Vidoptimus Image Server; before, turning it on switched your project to Runware and the style lock said your style picture would be skipped.
The server now stays selected with all three sizes available, and it receives your cast's faces and style picture, up to four pictures per image.
Export to CapCut
New: send your whole edit to CapCut as a project you can keep editing, instead of exporting a finished MP4.
Click "To CapCut" on the timeline; the button only appears if CapCut is installed on your computer (Windows or Mac).
Every picture, video clip, voice take, music track and sound effect goes on its own CapCut track, starting on the exact frame it plays at in the export, so voice and pictures stay in sync.
Transitions, slow zooms (Ken Burns), light effects and caption styling are left out so you can add your own in CapCut, and this doesn't change any timing.
Captions are saved next to the CapCut project as an .srt file, ready to import and style in CapCut.
Charts, text cards and other animated clips hold their last frame for the rest of their shot, just like the export.
A shot with no picture is filled with black, so the picture track has no gaps and later pictures never drift out of step with the voice.
Turn on "Copy media" to put copies of all your files inside the CapCut project; it's slower and uses more disk space, but the project keeps working even if you move your files.
Close CapCut before you export, because CapCut rewrites its project list when it closes, which can drop the new project; if that happens, just export again.
Exporting the same video again updates its CapCut project instead of creating a second one.
Fixes & Maps
Fixed: the timeline preview now shows the true shape of your video; before, a 16:9 video looked like a wide, short strip that cut off the top and bottom of every picture.
The live preview, generated preview and pre-render layout now all show the same 16:9 or 9:16 frame, sized to fit the panel, with text and layers exactly where they'll be in the export.
New map style: Relief shows real mountains, deserts and valleys under the borders, like a classroom atlas, and is ideal for wars, empires and journeys.
Choose Relief under Map style for the whole video, or give a single map its own style in its editor.
On a Relief map, the highlighted country rises off the map with its terrain still visible, tinted in your colour.
Globe and dive maps use Relief's colours, without the terrain.
Relief maps draw as fast as the other styles, and the terrain comes with the app, so no internet connection is needed.

v2.3.0 27 Sep 2026

Pictures now land on the words they belong to. The AI Director places every shot on the word it starts at, instead of giving it a share of the scene, so a picture of the harbour comes on screen as "harbour" is said, not a second before or after.
Cuts follow the voice itself: each one lands just before its word, however fast or slow the voice reads. Re-voicing a scene moves every cut with the new take, with nothing to plan again.
Editing a scene's narration keeps each shot on the same words. A line rewritten from scratch goes back to the old way of cutting until it is planned again.
New Plan step between Voice and Visuals: Script, Scenes, Voice, Plan, Visuals, Timeline. The Scenes screen now only splits the script and edits the lines; the shots are planned on the Plan screen, after the voice, where the timing is exact.
The Plan screen shows every scene's shots with the words each one is on screen for, and warns you when scenes are not voiced yet, since their cuts would be placed on a guess.
In auto mode, the AI Director plans scenes as soon as they are voiced, four at a time, and their pictures start filling while the rest of the script is still being voiced.
The Visuals screen quotes the words under each shot. On the timeline, the cut inspector shows where a cut falls in the narration, with buttons to move it one word earlier or later.
A long shot that needs splitting is split where the sentence turns — after a comma or a full stop, or on a word like "and" or "when" — instead of in the middle of a phrase.
Maps, charts and timelines stay on screen until their last place, figure or date has been said. A timeline no longer cuts away before its last year, and a chart of three years no longer shows only the first.
Maps, charts and timelines stay up for at least 3 seconds and text cards for at least 2, so they have time to animate and be read.
A text card that only repeats the number on the chart next to it is left out.
A chart or timeline now appears on its figure, not a word or two before it.
Each cut rhythm now has its own look, not only its own speed. Documentary opens on wide shots, prefers real footage and puts places on maps and dates on timelines; Calm uses spacious pictures and gentle motion; Standard mixes wide, medium and close shots; Punchy favours close-ups and puts key lines on cards.
Pictures also move to match the rhythm: a slow drift in Documentary and Calm, a stronger push-in in Punchy. Standard moves as before.
Every image style now has its own prompt. A third of them, Nature Documentary among them, were using one generic line and looked much the same. Each style also has its own description in the picker.
Two new image styles: Islamic Art and God Anime Vine.
New: your own image styles. Under Custom in the style picker, click "Add custom style", give it a name and describe the look in your own words — colours, light, texture, mood — and it is kept for every project, next to the built-in styles. Hover a custom style to edit or delete it.
"Generate preview" draws a sample picture in your style with your image provider, so you can check the words before making a whole video with them.
Your style's words go with every AI image as its style, and the AI Director plans its shots for that look. A project keeps its own copy of the style, so editing or deleting it later does not change a video already made with it.
Text cards are much bigger and bolder, sized to the frame instead of to the captions, and never run off it. Lines are balanced, so a card no longer ends on a single word, and flow and list cards are easier to read.
Opening hooks are set as headlines: the punch line and the words before it are much bigger, the picture is darker behind the words so they read over a bright sky, and the glow matches your accent colour. The Cold open's line, the Redacted file's revealed line and the Countdown's label are now large enough to read.
The Glitch hook's punch line is always the bigger line, and a long one wraps onto two lines.
Cards and hooks are redrawn in the new look the next time you render.
In an auto run, a stock file that can't be used — over 250 MB, removed, or in a format the app can't play — no longer fails the scene. The next best result is tried instead, up to four in all.
Vid Optimus now tells you when a new version is out. On Windows it downloads in the background, and one click on "Restart to update" saves your work, installs it and opens the app again — never in the middle of a render. On macOS and Linux a notice in the corner has a Download button.
Settings has a new Updates section with your version and a "Check for updates" button.

v2.2.0 25 Sep 2026

New: an opening hook. The first seconds of the video grab the viewer — words from the narration build over the first picture, and a punch line slams in on a hit, with a punch-in, a shake, a flash and its own riser and impact sounds.
Eleven hook styles: Statement, Number (a counter races up to a figure), Question, Highlighter, Focus reveal, Cold open (black, a flash, the date), Redacted file, Countdown (a red clock runs down), Glitch, Text knockout (the picture shows through giant letters) and Rapid fire (three hard cuts).
It is your choice, like motion graphics: tick "Opening hook" in the AI Director settings or the new-project wizard, and pick which styles it may use — or leave them all open and the Director picks the one that fits your opening line.
A hook only ever shows words the narration says. A number, date or phrase that isn't in the script is never put on screen; if the Director's idea doesn't pass, the hook quotes your opening line instead, or is left out.
Any project can have a hook, Director or not: the new Opening hook panel at the top of the Visuals screen switches it on or off, changes the style, lets you edit the words (with a warning if they're not in the narration), and can suggest another.
The hook plays over the first shot's own picture. On a still, the picture then carries on with a slow push-in; on a stock video, the video carries on from where the hook left it. The preview and the export look the same.
The first shot is given four seconds for the hook when its scene has room; on a shorter shot the hook plays faster rather than being cut off. Videos without a hook cut exactly as before.

v2.1.1 25 Sep 2026

Together AI pictures work again. Its old FLUX.1 models no longer answer, so the Together list now offers FLUX.2 Dev, FLUX.2 Pro, Juggernaut Pro Flux and Juggernaut Lightning Flux. Pick one of these if a project still has an old Together model selected.
Each Together model shows its price in the picker, and the Visuals screen estimates what a run will cost: about $0.002 an image for Juggernaut Lightning Flux, $0.005 for Juggernaut Pro Flux, $0.015 for FLUX.2 Dev and $0.03 for FLUX.2 Pro.
When Together turns a model away because your account has third-party data sharing switched off, the app now says so and tells you where to switch it on, instead of showing a raw error.
When Together is busy and asks the app to slow down, scenes now wait their turn and try again, instead of failing with "together 429". A scene only gives up if Together is still refusing after five minutes, and then it says so in plain words.

v2.1.0 24 Sep 2026

Background music now ducks under the voice: it dips while the narration speaks and comes back up in the pauses. It is on for every project, with a new "Duck under voice" slider next to the music volume. The volume slider now sets how loud the music is in the pauses.
The finished video is levelled to −14 LUFS, the loudness YouTube and TikTok play at, so it is neither quieter than the videos around it nor turned down by the platform. It is on by default; switch "Level for YouTube & TikTok" off under Music to keep the old mix.
In the Documentary rhythm, maps, charts and timelines stay up for at least 5 seconds and at most 12, while they are being spoken about. When the narration talks about a map for longer than that, the next picture takes over, and the AI Director is told to plan one.
A map, chart, timeline or text card held after its animation finishes now pushes in slowly until the cut, instead of sitting still. The preview does the same.
In the Documentary rhythm, a picture held for longer than 4 seconds now moves further, so its Ken Burns move keeps its pace instead of slowing to a crawl. Calm, Standard and Punchy pictures move exactly as before.

v2.0.0 23 Sep 2026

Maps look new: countries trace their outline, rise off the map with a soft shadow and a gentle glow, and the map around them dims so the place you are talking about stands out.
Pins drop in with a pulsing beacon, and labels sit on a pointer so they never cover the place they name.
Globes show stars behind them and real shading on the sphere, and a dive from space lands on the place much more smoothly.
Four map styles — Clean, Dark neon, Vintage atlas and Blueprint. Pick one for the whole video under Map style, with a live preview, or give a single map its own style in its editor.
The Vintage atlas style uses a serif font, paper grain and a darker border, like an old printed atlas.
A route can now carry a plane, a ship, a car or a train along its line. The AI Director picks one when the narration says how the journey was made, and you can change it in the map editor.
Four new kinds of map. Campaign arrows show armies or movements pushing across the map. True size lays one country over another at its real scale. Bubbles size a circle on each place by the number the narration gives. Distance draws a line between two places with the distance on a badge.
The AI Director knows about every new map kind, and only uses one when the narration actually gives it the places and figures it needs. Otherwise it falls back to a simpler map, just as before.
Maps draw much faster — a flat map takes about half the time it did — and the app no longer freezes while a map, chart or timeline is being drawn. Redraw stays responsive, and the Visuals and Timeline screens keep scrolling while graphics are made.
New Documentary cut rhythm for the AI Director: pictures hold 5–7 seconds instead of at most 4, about 11 cuts a minute, and maps and charts still stay up while they are being spoken about.
The preview, the timeline, sound effects and the export all follow the Documentary rhythm, so cuts land in the same place everywhere.
Calm, Standard and Punchy cut exactly as before, and projects that do not use the AI Director are unchanged.
The four cut rhythms sit in one row in the Director card and in the new-project wizard, and the wizard's review step shows the rhythm by name.

v0.1.9 23 Sep 2026

Generate all on the Visuals screen no longer closes the app on long projects. Stock clips are now saved straight to disk as they download, instead of being held whole in memory, three at a time.
Stock files over 250 MB are refused with a message asking for another pick, on every source — the Internet Archive was the only one with a limit before.
Runs that fetch stock video download two clips at a time instead of three, which is easier on slow connections and disks.
Wikimedia Commons clips download again — every Commons pick was failing with "download returned 403".
The Visuals screen shows small copies of each picture and only loads the ones on or near the screen, so a long project scrolls lightly instead of running the app out of memory.
A reference photo too large to read safely is refused with a message instead of taking the app down.
Generate all on the Voiceover screen and on the Visuals screen can now run at the same time — starting one no longer greys out the other until it finishes. Stop in the progress bar stops both.
An unexpected error now shows a recovery screen with Try again and Reload, instead of closing the app or leaving a blank window. Your project is saved as you work, so reloading loses nothing.
Errors are written to a log file, so a crash report can say what happened. On Windows it is in %LOCALAPPDATA%\com.vidoptimus.app\logs.

v0.1.8 22 Sep 2026

Free-tier keys no longer break a run: when Groq, Gemini or OpenAI answers "too many requests", the app now waits the time the provider asks for and carries on instead of failing.
The AI Director no longer hands back a half-planned edit on a rate-limited key — scenes that used to come back silently unplanned are now either planned properly or reported as a rate limit.
A rate limit that turns out to be the day's quota rather than a short burst is reported straight away, so a run says what is wrong instead of sitting there for an hour.
Whisper captions survive a rate limit too: a long transcription keeps every part it has already done instead of throwing the whole recording away.
Generate all on the Voiceover screen now picks up where it stopped, voicing only the scenes still missing instead of starting again from scene one.
A scene that has its narration but lost its captions is repaired with a single Whisper call, without paying to speak it again.
Voicing a whole project now reports into the progress bar at the top and can be cancelled mid-run, the same way Visuals already did.
Google Gemini added as a language model, with a free tier that needs no credit card.
Gemini models are labelled with their real daily free limits, so you can see which one will finish your script before you start it.
OpenAI's GPT-5.6 Luna added to the model picker.
Reasoning models are now given room to think on top of the answer they were asked for, fixing prompts and shot plans that were being silently cut off mid-answer.

v0.1.7 22 Sep 2026

Stock footage now searches NASA, Openverse, Wikimedia Commons and the Internet Archive — four open archives that need no API key at all, so a fresh install can fetch clips before you have pasted anything.
A search fans out across every source that is ready, so one archive being down costs only its own results and nothing more.
Results from different sources are interleaved rather than stacked, so no single archive dominates the picker.
NASA's plain http links are upgraded to https rather than downloaded in the clear.
Oversized Internet Archive items — whole films, measured at up to 7.8 GB — are dropped rather than downgraded, and the source is kept out of unattended runs.
Wikimedia Commons falls back to an API gateway on networks that block commons.wikimedia.org, and those results are marked as needing a credit rather than guessed at.
Seven ready-to-paste demo narration scripts added, including one written to exercise the free archives with no stock key saved at all.

v0.1.6 22 Sep 2026

Urdu and every other right-to-left script now lays out the way it reads — highlighted captions used to come out with their words scrambled.
Per-word highlighting, and the pop, ghost and karaoke caption styles, all render correctly in right-to-left scripts.
Right-to-left captions break their own lines in spoken order, since a mirrored line cannot be wrapped automatically.
Motion graphics mirror their placement, alignment, bulleted lists, wipes and caret in right-to-left, while the typewriter, scramble and emphasis effects still run in reading order.
Left-to-right captions are byte-for-byte unchanged.

v0.1.5 21 Sep 2026

The voice picker lists all six AI33 back-ends rather than ElevenLabs alone — a MiniMax or Vbee voice speaks through the same call.
Each voice carries its engine in the label, and a long list can be filtered by name, engine, language or gender.
A dropped voices page is retried, then re-asked in smaller slices, and a sweep that collects anything at all no longer reports an error — one bad page used to empty the whole picker.
The saved voice catalogue is merged rather than replaced, so a short sweep never shrinks it.

v1.0.0 17 Sep 2026

First release as Vid Optimus.
Paste a script and get a finished video: AI Director plans the shots, narration and visuals generate together in Auto mode.
Three visual sources per scene — AI images (5 provider integrations, 40+ models), stock photos and stock video.
Consistent characters: keep a recurring face the same across shots, opt-in.
Word-level captions snapped to your own script.
Full timeline editor — Ken Burns, transitions, fonts, live 60fps preview, continuous autosave.
Frame-accurate MP4 export in 9:16 or 16:9, with automatic hardware-encoder detection.
Windows and macOS, one licence for both, one device active at a time.
API keys stored encrypted in your system's own secure keychain.

What's new in Vid Optimus

New versions arrive as normal updates.