AI Director, AI Designer and media generation
Build or refine projects conversationally while keeping every accepted result editable.
JsonCut separates project direction from media generation.
| Tool | Scope |
|---|---|
| AI Director | Discusses, builds and refines a video project, including layers, scenes, motion and media |
| AI Designer | The same Agent for image and carousel projects: builds and refines the design, layers and media |
| Media generation | Creates images, video clips, voiceover, music or sound effects for the current project |
Review before accepting
An AI response is not the final artifact. Ask for a bounded change, inspect the result on the canvas, then keep editing the layers. Edits go onto the canvas directly, in videos and in images. Every run saves the state before it to the version history (clock icon in the top bar), so you can go back to it; Rollback on an applied edit in the chat does the same for that edit.
Steering the AI Director
- Choose Lite or Pro in the director menu of the message box. Lite is fast; Pro uses the strongest model, is billed by usage and is part of the Pro and Business plans. The Effort bar in the same menu sets how long it thinks and checks, and Look picks a visual style (or lets the Agent choose).
- Right-click a layer and choose Ask AI Director… (Ask AI Designer… in images) to start a message about exactly that layer, with suggestions that fit it.
- Selected layers and, if you want, the moment at the playhead appear as chips above the message.
- Before / After on an applied edit shows the state before it in the preview; changed layers glow briefly in the timeline and the layer list.
- Save as shortcut keeps a message you use often; type
/in the message box to reuse it. Studio also offers ready-made shortcuts such as "Clean up talking head". - Changes made through MCP or the API appear in an open Studio within a few seconds.
Generated media
Generation operations are asynchronous. Depending on the selected operation, Studio accepts a prompt plus bounded controls such as aspect ratio, duration, voice, pace or variant count. Generated results appear as project media and are not hidden inside the Agent message.
Images and video. Under Generate, pick what to do first: Create or Edit a picture; for video Text, Image (from a start image), Edit or Avatar (talking head). The card at the top shows the chosen model; Change opens all models that can do this, with their maker, strengths and price in credits (Nano Banana, GPT Image, FLUX, Seedream, Grok, Veo, Kling, Seedance, LTX, PixVerse and more). Recraft V4.1 Vector makes icons, logos and illustrations as SVG files that stay sharp at any size. The controls below offer only what the model can make: formats, resolution, variants, length and sound. New pictures and videos start in the project's format; an edited picture keeps its own unless you choose another, and a video from an image or an edited video keeps the format of its source.
- Reference pictures (Create) guide subject and style; in Edit, Picture to edit is the one that changes and further references guide it. Upload them or pick files from the project.
- Background: Transparent (GPT Image 2.6 and 2.5, Sunburst and Flare) makes a picture without background, for stickers, cutouts and product shots on any layout.
- Image animates a start image; Edit changes an existing clip (its length is the result's and the price).
- Avatar lets a portrait speak a voice track, for example a voiceover from Audio → Voice. The video is as long as the voice and keeps the portrait's format; a prompt is optional.
- Results appear as cards: hover shows the details, a click opens a large preview, + adds the file and you can drag it onto the canvas or timeline. While a model works, its card shows the progress, estimated from the model's usual time.
The AI Director and the AI Designer generate media themselves when a brief asks for it. By default they use Nano Banana 2.1 for pictures and picture edits, Grok Imagine Video 1.5 for clips, Gemini Omni Flash 1.1 for clip edits, VEED Fabric 1.0 for talking heads, Stable Audio 3 for music and ElevenLabs SFX 3 for sound effects; a picture without background uses GPT Image 2.6 Sunburst. They wait for the file and place it in the project, for example a generated product photo as the start image of a clip with titles on top. The options button (sliders) at the top of their panel decides what they may generate at all and which model each task uses; with Generate media with AI off they work with stock, your files and editable layers only.
Music and sound work the same way: the card shows the chosen model and Change lists them all, each with a short example of the same brief to play. For music: Stable Audio 3 (the default: fast instrumental beds for a few cents), Eleven Music 2.5 (exactly as long as you set, with or without vocals), Mureka 9.5 (the best-rated songs, vocals from lyrics), Lyria 3.5 and MiniMax Music 3. For sound effects: ElevenLabs SFX 3 (loops without a gap, or a length it chooses itself), Mirelo SFX 2.0 (foley), Stable Audio 3 SFX (long ambience) and Sonilo SFX. The options follow the model: a length slider, or a whole track to trim on the timeline; vocals and lyrics only where the model sings; loops only where it makes them. Every generation shows its credits on the Generate button before it starts; failed or cancelled generations release them.
Speech in another language. Translate speech in Audio → Voice translates the spoken lines of a video or audio clip into another language and speaks them with a voice you choose. You can change every line before it is added; each line is placed where it was said, and the original clip can be muted, with its captions replaced. The AI Director can do the same when you ask for it. Details are in Audio and captions.
Productive prompts
Describe the outcome and the part that may change:
Shorten the opening to two seconds, keep the product shot and brand palette, and add a clearer final call to action.
For a new visual direction that spans multiple turns, you can optionally use a cut.md Agent brief so audience, locked rules, media policy and quality checks remain explicit. It is not required for Studio editing or export.