What problem does it solve? Turning a natural-language image request into a valid ComfyUI workflow graph normally requires hand-editing node JSON. This Skill deterministically compiles a Visual Intent Object (VIO) into a ready-to-submit ComfyUI prompt JSON by patching a bundled template, with no LLM involved in graph generation. ## Core Features & Use Cases - Template-based compilation: Ships SDXL and FLUX.1 txt2img templates; the correct one is auto-selected from the model checkpoint hint. - Flat override mapping: Maps VIO fields (prompt, negative prompt, width, height, steps, cfg, sampler, scheduler, seed, checkpoint) onto the correct node inputs. - Defensive validation: Refuses to run with a missing or empty prompt instead of silently falling back to the template's placeholder text; emits structured error JSON with exit code 2. - Use Case: A planner emits a step with goal=generate and a prompt like "a blue cat" with steps=30 and seed=42; pipe the VIO into compile.py and feed the resulting prompt JSON directly to a ComfyUI submission script. ## Quick Start Pipe a Visual Intent Object JSON containing your prompt and parameters into scripts/compile.py and send the stdout prompt JSON to the ComfyUI submission script.
