Image generation
OpenAI and fal.ai image studio, jobs, history, and agent tools.
Image Gen is a project-bound Tool node backed by a native async runtime. Generation continues even if the canvas window is not focused; only opening the node uses the UI bridge.
Providers
| Provider | Role |
|---|---|
| OpenAI | Fixed gpt-image-2 generation and source-conditioned edits; size, quality, PNG/JPEG/WebP, compression |
| fal.ai | Catalog text-to-image / image-editing models with validated schemas |
Configure keys under Settings → Connections (OpenAI, fal). Keys are native keyring vault entries.
Prompt assist
With OpenRouter configured, the sparkles assistant can enhance or rewrite a prompt using your Global Agent model (no tools). Undo is available via toast. Manual prompts work without OpenRouter.
Jobs and history
- Start returns an opaque job reference immediately
- States: queued → running → succeeded / failed / cancelled / interrupted
- One app-wide paid job slot at a time
- OpenAI jobs are not provider-resumable after restart
- fal queue jobs can resume unfinished work after restart when ids remain valid
- Assets stay in private project storage with retention rules; publish is no-overwrite
Terminal outcomes also feed the Notification Center.
Agent tools
Global Agent and workspace Agent CLI share catalog, start/read/cancel, and publish tools. External MCP has no Image Gen tools. Results stay metadata / opaque refs — not raw credentials or free-form host paths in IPC.
Canvas
- Singleton studio node per project
- Optional reusable Image Helper for viewing generated media
- Screenshots and layout follow general node rules
Tips
- Connect providers before first generate.
- Prefer cancel over force-quit during long fal jobs if you care about clean state.
- Check Notifications for background failures when the node is closed.