tools / image_gen.py

Stable Diffusion XL image generation routed to an external GPU node.

Tool: image_generation

Auto-registered when context.llm_client.image_gen_clients exists.

• Heals missing prompt by checking related kwargs (image, description, subject, text).
• Clamps steps to [4, 8] (Lightning models hate longer schedules).
• Calls llm_client.generate_image({"prompt", "steps"}).
• Decodes base64 PNG, writes to sandbox as gen_UUID.png.
• Returns markdown embed: ![prompt](/api/download/gen_UUID.png)

Backend topology and SDXL model config: image-gen server.