Changelog

Comics and dramas get a BGM sheet: play the score from the card, and swap it on its own

The score never had a place of its own. A track generated while rendering the cut was just another audio file on the work — nothing on the page would play it, and nothing recorded what style it was made to. So the next episode got a freshly generated track, nobody could say why it sounded different, and when the music felt off all you could do was half-hear it under the narration and describe it back. Reference sheets now include a "BGM sheet", sitting next to the voice sheets: the card describes what this work's score is like (genre, instruments, whether it stays under the dialogue), and the track attached to it is the one actually used in the cut — open the card and play it. What the card holds is one short loop that joins back to itself (fifteen to twenty-odd seconds), not a whole song: episodes differ in length, so rendering just tiles it to fit. Storing a full song instead means a short episode never reaches its second half, and a long one restarts the intro in the middle of a scene. If it is not right, swap in another track on the same card; not a single panel or line of voice-over has to change. Sticker packs are still images, so they do not have this.

A work now keeps only what still applies: series-level story split from per-episode, and voice sheets you can play

A work's notes used to collect the story of how it got here — "was written as seven heads tall, corrected", "user confirmed:" — which reads like a chat log, while what you open a work for is what the work is now. That column now holds only rules that currently apply: when something was decided more than once, you see the last version, and character or art-style settings are separated from production numbers like resolution, frame rate and subtitle styling into "Creative rules" and "Production specs". The "Story" block is now "Series story (whole run)" and holds only what stays true across the whole run; what a given episode is about, and how its emotional arc moves, belong to that episode's own outline. Previously episode 1's plot got written up as the story of the entire series — and a series that has just started genuinely has no series-wide storyline yet, so leaving it empty is the right answer. "Open items" is gone as well: a work only ever grows — more episodes, more sets — so there is no finish line and nothing is outstanding; deciding a panel is fine as it is makes it a creative rule. Voice sheets now carry a sample clip you can play from the card — no amount of reading "a bright young man's voice" tells you whether it is still the same voice — and retired sheets get deleted outright, instead of leaving a card marked "deprecated" for you to second-guess — you can delete one yourself too, from "Delete this sheet" inside the card.

New image and video models, and one quota split into two

The models behind image and video generation have both moved up a generation: images now run on image 2.5, video on seedance 2.5. From the same description you get a picture that sits closer to what you asked for, with steadier detail and steadier text inside the frame; when you ask for an edit, the parts you did not point at are less likely to move along with it; and in video, motion and cuts hold together better, so a few seconds of footage reads as filmed rather than assembled. Separately, what used to be one quota is now two, counted independently. Main model usage · every 5h covers chatting, writing code, changing apps and producing documents, measured over a rolling five-hour window that refills on its own. Tool usage · weekly covers image generation and editing, video, voice-over and transcription, measured over the calendar week and restored in full every Monday at 00:00 UTC. Before, the two shared a single pool, so a run of a few dozen images could throttle ordinary conversation as well; now running one down leaves the other untouched, and the meter in your profile shows two rows, one percentage each.

Works now show a material checklist instead of progress, plus contents bars, episode pickers and codes

Works gained two categories — AI short dramas and sticker packs — sharing the same reference sheets, codes and change-one-thing-without-redoing-the-rest as animated comics. The progress line on a work (which used to read “Not started”) is gone: a work only ever grows, so there is no moment where it is done. In its place is a material checklist that simply says what exists and what does not — 3 character sheets, episodes 1-2 fully drawn, 4 panels still missing in episode 3, no voice-over yet — and you and the agent read the same one. Each tab now opens with a contents bar (how many character, scene and style sheets there are, one click to jump), and the storyboard shows one episode at a time from a dropdown, each entry carrying its own coverage (“Episode 3 (18/20)”), so a work with dozens of episodes no longer becomes a page you cannot scroll to the end of. Every panel and every sheet also carries a code — EP3-P5 for episode 3 panel 5, SET2-S7 for sticker 7 of pack 2, CHAR-Mika for a character sheet. Say the code and the agent knows exactly which one you mean, with no counting involved; there is also an “Ask AI to change this” button that opens a new chat with the code attached. Titles, summaries and covers are editable now too — open the work and use Edit in the top right.

Storyboards grouped by episode, one character per sheet

The storyboard tab no longer spreads every panel's text across the screen. Each episode is now a group of thumbnails, so you can see at a glance how it turned out, and clicking a panel opens that panel's details — the shot, the lines, what it has to get across — with next and previous to page through them. When there are a lot of episodes, click a heading to fold away the ones you have already been through. Character sheets have changed too: one sheet now holds exactly one character. If the reference you give is a group photo or a still with several people in it, the agent cuts your character out first instead of treating the whole photo as the character design — otherwise every later panel follows that image and drags the other person into the frame.

Works get cards and a detail page split by section

The works list now looks like the app list: a row of cards with a cover, the title, the category and a line of progress such as “Step 3/7 · Character sheets”. You do not upload the cover — the first design plate you draw becomes it. Inside, the tabs run down the left in a single column (they used to squeeze into a grid of small squares), and the content is no longer one long block of text: the premise, the story, each episode and each panel all get their own place, character sheets are a wall of cards with their art, storyboard panels carry thumbnails, and the finished film plays right on the page. Which fields a work has is settled the moment you pick its category, so they are a fixed set rather than something that varies from one work to the next.

Works: a home for dynamic comics and other long projects

Your profile has a new Works entry, sitting alongside apps. A dynamic comic runs to hundreds of panels, several episodes and a whole set of character designs; until now all of that lived inside one chat, and opening a new chat meant it was out of reach. It now lives in the work itself, which belongs to your account rather than to a single conversation. A work is split into outline, character sheets, storyboard and the finished film, and a character sheet keeps not just the reference art but a paragraph that pins the character down in words — which is what keeps the lead in episode seven the same person as in episode one. Production runs outline → designs → storyboard → panels → voice-over → assembly, and the first three steps each stop to check with you: changing things early is cheap, and finding out the character is wrong after a hundred panels are drawn is not. When you want to change a design you are told what it costs first (“6 panels get redrawn, the other 14 stay”), and nothing moves until you agree, so shots you already approved are never lost. To pick a work back up, @ it from any chat. For a couple of images or one short clip there is no need to create a work — just ask in the chat.

Voice-over, transcription, and @ mentions that carry a type

The agent can read your text aloud now — narration, spoken promos, audiobooks, the lines in a dynamic comic. You can pick a voice, and you can describe the delivery in a sentence (“gentle tone, a little slower”). The other direction takes one sentence too: hand it a recording and ask for a transcript or subtitles. Separately, when you @ something, the input box now writes it as “@[Work]{name}” with the type spelled out, so it stays clear to you later and the agent no longer has to guess from the name whether you meant a file, an app or a work.

Interactive reports open fully expanded

Opening an interactive report now shows every section expanded. Appendix and methodology notes that used to start collapsed are visible immediately, so you no longer miss them thinking they were empty. You can still collapse a section or use Collapse all when you want to.

Suggested prompts live in the report

The first message of a scheduled run no longer shows a row of quote marks. The agent also no longer pastes the full suggested prompt into the chat bubble — when the run produces an interactive report, that section goes at the end of the report with a copy button, and the reply stays a couple of sentences.

Scheduled task prompt review

Creating a scheduled task, or rewriting an existing one's prompt, now opens a chat that reviews it with you. Anything where a different answer visibly changes what you get goes on one card — who “my favourite actor” actually is, how far back to look, how many items a day, what shape the output takes — while what only affects how it works (file naming, retry counts) is settled without bothering you. Every question can be left blank to take its recommendation. Once you have answered it hands back a complete prompt you can paste straight back. Saving no longer drops you into that chat: a dialog explains what the review is and asks whether you want to open it, and Not now keeps you on the list. A new Prompt review column shows the chat id of the most recent review and the status it was in, so you can go back to it whenever you like. When you paste the reviewed version back in, tick “This prompt has already been reviewed” under the prompt box and it will not be reviewed a second time. Runs no longer stop to ask you either — nobody answers at 3am, so the agent proceeds on the most reasonable defaults and closes the report with an improved prompt you can swap in over breakfast.

Scheduled tasks

Schedule in the sidebar lets you register a prompt and a time once, then have it run every day or every week. When the time comes we start a new chat for you and send that prompt in verbatim, with attachments and @ mentions working exactly as in the normal input box. Tasks can be paused and resumed, and the execution history links straight back to each chat they produced.

Cloud drive

Cloud drive in your profile collects every file you generated or uploaded across all your chats, newest first and ready to download, so you no longer have to dig back into the original chat to find one.

Changelog

There is now a place to see what changed in each update. Open it from Changelog in the top navigation or the footer before you sign in, or from the last entry in your profile menu after you sign in.

Browse showcases without signing in

Use cases, the app gallery, the image gallery, the video gallery and pricing all open without an account, so you can send the links to anyone and search engines can index them. Entry points are in the top navigation and the footer of the home page.

Five-hour quota meter

Your profile now shows how much of the current five-hour quota you have used and how much is left. The bar changes colour as you approach the limit, so you find out before you get throttled.

Ship your app as an Android app

Ask for an Android build in the chat and the agent sets up the build, produces the APK, and adds both a download button and a QR code to your app website.

Monitoring board for every app

Each app has its own board covering traffic, referral channels, frontend errors and search-engine crawling. When monitoring data is missing, the board offers to wire it up for you.

Marketing profiles

Fill in your positioning, audience and tone once under Marketing in your profile, and the agent reuses it every time it drafts posts, so you never have to explain the background again.

Terms and policies page

The terms of service, refund policy and privacy policy now live on one page that opens without signing in, linked from the home page footer and from pricing.

Sign in with Google

The sign-in dialog now offers Google, so you can get in without keeping track of another password.

Share your work to Explore

Apps, images, videos and other output from a chat can be shared to Explore under four categories, and the resulting link opens for anyone without signing in.