DocsGenerative AI and Flows
Generative AI and Flows
Generate images and 3D models with your own provider keys, keep them in a local library, and chain steps such as sketch, character sheet and 3D model into a Generation Flow.
Connect a provider#
Generation uses its own keys, separate from the chat's accounts. Add them in Preferences ▸ AI Providers ▸ Generative AI (API keys). If you are in a team, its generation keys appear too.
| Connection | Makes | Models |
|---|---|---|
| OpenAI Images | Images | GPT Image 2.5 Flare, GPT Image 2.5 Sunburst |
| Gemini Images | Images | Nano Banana 2.1, Nano Banana 2, Nano Banana Pro |
| Seedream Images (BytePlus) | Images | Seedream 5.0 Pro |
| Tripo 3D | 3D models | Tripo P2, Tripo 3.1 |
| Meshy 3D | 3D models | Meshy 7, Meshy 7.1 |
Nano Banana 2.1 appears when your Gemini key offers it; otherwise Nano Banana 2 is the default.
The Generative AI editor#
In Manual mode, open any area's Editor Type menu and choose AI ▸ Generative AI. In Director, just ask in the chat: the agent can generate for you without the editor.
- Choose Image or 3D, write a prompt, and pick the model and its settings: aspect ratio and resolution for images, a target polygon count and textures for 3D.
- Add up to four reference images with Image, URL or by dropping them on the editor. A 3D model is made from a text prompt or one reference image.
- AI Enhance rewrites your prompt; you review the result before it replaces yours.
- Generate image or Generate 3D shows the model, the account it bills and what will be uploaded, and runs when you confirm. Up to four generations run at once, and nothing is retried automatically.
Stopping a generation only stops Angelo watching it: a task the provider already received may still finish and be billed.
Angelo shows no prices here; your provider bills you as usual. It makes images and 3D models only: not textures, video or audio.
The library#
Results go to Angelo's local library, shown in the Generative AI editor, with folders and search. Nothing is put in your scene by itself. Click a result to:
- Use as reference (images) for the next generation,
- Import into scene (3D models, as GLB), which Undo can take back,
- Export… a copy anywhere; an existing file is never overwritten.
The library lives in ~/Library/Application Support/Angelo/media-studio on a Mac and %LOCALAPPDATA%\Angelo\media-studio on Windows, shared by all your files. Import adds your own images to it.
Generation Flows#
A Generation Flow chains steps into one node tree, for example a sketch that becomes a character sheet, then a turnaround, then a 3D model placed in your scene. It uses the same models, keys and library.
- In a Node Editor, switch the tree type to Generation Flow, open the sidebar's Flow tab and click New Generation Flow. Or ask the agent to build one.
- Add nodes from the Add menu: Image, Scene Capture, Text, Generate Image, Generate 3D, Place in Scene and Subflow.
- Click Run flow, or Run this step or From here on a node. The Run Generation Flow dialog lists each paid step with its account and estimated price (≈) and the total; tick the steps you want and click Run.
Steps that didn't change reuse their earlier result for free. One run makes at most 16 paid generations; a Stop for review on a node splits a bigger flow into parts you approve one at a time. When the agent runs a flow, the same list appears as a card in the chat, and it always asks. A finished flow can be saved as a recipe and reused as a subflow.
Templates: a character in parts#
Coming soonA template is a ready-made Generation Flow. The first, Character in parts, turns one picture of a character into separate 3D parts (body, hair, headwear, clothing, footwear and accessories) placed together in your scene.
- In the Generative AI editor, choose Workflows. Pick From an image and drop a front view of the whole character, or From a description and describe them. Set the height.
- Make parts sheet asks first, then makes one image with every part laid out in columns. Angelo cuts it into parts on your computer, for free.
- Check the sheet: each part is outlined. Turn off any part you don't want, or Redo sheet for a new one (a new paid image).
- Make 3D for N parts opens the usual approval with one 3D model per part. Each part shows its progress, and a failed part can be retried on its own.
- Place in scene adds the parts as separate objects under one empty at the 3D cursor, in one step Undo takes back. Clothing sits just outside the body, and body hidden under clothing is masked, not deleted.
Treat the result as a starting point to finish in Blender. Hats, hair and accessories can land off: the editor lists them so you can check and move them. In the Node Editor, the same run is an ordinary flow you can open and change, and Add ▸ Templates inserts a template there directly.