Intent Routing
How perstudio turns natural language into the right ComfyUI workflow.
How it works
When you call POST /generate with an intent string, perstudio uses Claude (Sonnet) to analyze your request and match it to the best available workflow. The routing process:
- 1. Intent analysis — Claude parses your natural language to identify what type of image you want, the subject, style, and any technical requirements.
- 2. Workflow matching — The router compares your intent against all available workflows and their annotated capabilities (template_slots, categories).
- 3. Slot filling — Claude maps your intent into the workflow's template slots (prompt, negative_prompt, dimensions, etc.).
- 4. Execution — The patched workflow runs on the GPU pod and returns your generated assets.
Supported categories
Studio-quality product shots with controlled lighting
Photorealistic portraits, headshots, and group photos
Editorial art and illustrations (Illustrious-based)
Cartoon, anime, and manga-style art
Architectural visualization and interior design
Apply artistic styles from reference images
Structure-guided generation from control images
Transform existing images with prompts
4x resolution enhancement with detail recovery
Edit specific regions of an image using a mask
Die-cut stickers with transparent backgrounds
Text-to-video, image-to-video, and talking heads
Text-to-speech with voice cloning
Routing transparency
Every response from /generate includes the workflow_name and category fields so you can see which pipeline was selected. If the routing isn't choosing the workflow you want, you can use Tier 3 endpoints to target a specific workflow directly.