If you need to turn a square Instagram post into a 16:9 YouTube thumbnail or uncrop a portrait to show more body, this routes your request across RunComfy's identity-preserving edit endpoints. Nano Banana 2 Edit is the default for spatial language like "extend canvas left and right by 30%", GPT Image 2 Edit handles multi-reference style matching, and FLUX Kontext Pro does single-shot maximum preservation. The routing logic is smart about matching your ask to the right model, but if you hit visible seams or lighting mismatches at the boundary, you'll need to drop into a ComfyUI workflow for finer control. Works best for aspect ratio changes and moderate extensions, not pixel-perfect texture continuation.
npx skills add https://github.com/agentspace-so/runcomfy-agent-skills --skill image-outpainting