Eight tools, from the script to the final clip.

The script read

What does the AI script read do?

It reads the whole script and says what has to be made. Two reads run side by side: one for the cast, one for the world.

The cast read lists every character with a description to cast from, and the look they wear in each scene. The world read lists every location, the props the story names or implies, and which scenes run on without a cut.

• TakesA PDF, a Word file or pasted text
• Who reads itThe app lists the scenes itself. Claude, from Anthropic, reads the title and first pages to name and set up the project. The cast and world reads take the whole script, and are done by Claude or by your own connected LLM
• FindsScenes, characters and their looks, locations, props, continuity between scenes
• KeepsYour own decisions: what you dismissed or linked stays as you left it

Scenes

How are scenes organized?

A project holds scenes in script order, each with its heading and time of day. A scene holds one or more sequences, and a sequence holds the shots: it is a run of shots that render together as one clip, so a long scene is cut into several. A series or a vertical drama adds episodes above the scenes.

Each sequence stays tied to the script lines it covers, so the page is always beside the shots made from it.

• ProjectA short, a series or a feature, with its frame and its look set once
• SceneFrom the script, with what it needs and the sequences made for it
• SequenceA run of shots, rendered together as one part of video

Bibles

What is an automatic bible?

A bible is what stays true. Each character, place and prop has one, drafted from what the script says about them: who they are, what they look like, what they wear.

Each sequence has one too, in eight parts, written once and sent with every shot: general description, style, location, lighting, continuity, physics, sound and what never to show.

• DraftedFrom the script, by the app or by your own LLM
• VersionedEvery save keeps the version before it
• UsedSent at the top of every prompt, so no shot repeats it

The asset library

What is an asset, and what is an asset call?

An asset is something that must not change between shots: a character, a world or a prop. It is made once as pictures and given an @ID.

An asset call is typing that ID in a prompt. The asset's pictures are attached to the shot and sent as references with every still and every render of it. The model is shown the face; it is not told about it.

• CharactersA face row, face looks (make-up, a scar, a mask) and a wardrobe sheet for each costume
• WorldsA wide, empty still of each place, and its other sides
• PropsEach object alone on a studio backdrop, attached only where it appears

Shot controls

How do I control a shot?

With controls, not with words to remember. Each shot has its prompt, its dialogue, and a camera set by shot size, lens, move and angle. A floor map places the people and the camera, and warns when a shot breaks the 180-degree rule.

The tool shows its work. Review opens the whole prompt the sequence will send, shot by shot, before it goes, and can read it for ambiguity first.

• CameraSize, lens, move and angle, written into the prompt for you
• BlockingPeople, camera and the line of action on a floor map
• ReviewThe exact prompt, checked before anything is spent

Stills, video, upscale

How do pictures and video get made?

Frame a shot as a still first, for cents, on the model you choose. Then render the sequence as a whole: the video model gets the bible once and every shot in order, with all references, which is how one room stays one room across the cuts.

Draft at 480p, make the take you like final at 1080p, and upscale only the takes you keep. A draft is billed as 480p and its final at the 1080p price. Upscaling never renders again, so nothing in the shot changes.

Renders are downloaded as MOV or MP4 clips and cut together in your own editor.

• StillsGPT Image 2, GPT Image 2.5, Seedream 5.0, Seedream 4.5, Nano Banana 2, Nano Banana Pro
• VideoSeedance 2.5, Seedance 2.0, Seedance 2.0 Fast, Seedance 2.0 mini, Veo 3.1, Veo 3.1 Fast, Veo 3.1 Lite. A sequence opens on Seedance 2.5: 4 to 30 seconds a part, with sound. The engine menu offers the others
• Upscale1080p, 2K or 4K, priced by the second

Generation records

Is there a record of what was generated?

Yes, of everything. Each picture and each render keeps its full prompt, the model, the settings that were sent, its references and when it was made.

Any of them can be made again as it was, or changed by rewriting one line. The billing history shows what each job was charged and who ran it.

• KeptPrompt, model, settings, references, date
• ReusedReuse the settings, use the picture as an input, or save it as an asset
• CostedEvery job in the history with its price and the person who ran it, and a Budget page for each project: its spend by scene, sequence, person and model

Your own LLM

Where does an LLM speed the work up?

At every step that is writing. An LLM is an AI assistant such as Claude or ChatGPT. Connected to a project, yours reads the script, drafts the bibles, writes casting options, cuts a scene into shots, sets the camera and the blocking, and queues stills and renders.

It is optional: everything works without it. If you connect one, you talk to the project in plain words and approve what costs money. The app teaches the LLM its own craft rules, so you do not paste a manual into the chat.

• Connects byMCP, the open standard for giving an AI assistant tools, over HTTP with a key you make in the app. Set-up steps are given for Claude Code, the Claude app, Codex and Cursor, and for any other MCP client
• ScopedOnly the projects you tick, for 8 hours to 30 days
• PermissionsReading always; writing shots, queueing pictures and queueing video each switched on by you
• ApproveEvery paid proposal is a priced line that waits for a person