Current boundary

• Supported generators: map and frame-by-frame still generation with frames.
• Supported output: PNG images returned by one output node.
• Supported dimensions: 16–4096px per edge, subject to the workflow, model, sampler, VRAM, and node constraints.
• Provider generation may exceed the local refiner's four-million-pixel input limit. Crop or split those canvases before running refine.
• PixelKiln styleImages and palette are rejected. Put image references, palette, and shared controls inside the committed workflow. Use custom bindings plus providerInputs for per-asset ControlNet or reference images.
• Video models, persisted GIF/APNG preview files, multiple output nodes, arbitrary binary uploads, and ComfyUI Cloud authentication are not implemented. frames is a set of ordinary still prompts, not temporal sampling. Custom image bindings and revision inputs support manifest-owned PNG/JPEG uploads.
• PixelKiln does not install checkpoints or custom nodes. Every machine running the project must provide the models and nodes named by the workflow. The benchmark workflows use only core ComfyUI nodes.
• Background removal remains inside the committed ComfyUI graph. refine automates native-grid recovery, final palette enforcement, and measurable checks. A named human still approves the art at 1×; PixelKiln records and verifies that decision rather than making it.

For large mountains, buildings, and backgrounds, model choice and working resolution matter more than the provider label. The SDXL benchmark produced a coherent 384px cliff fortress and two layered environments where the starter SD1.5 workflow did not. Those results still required manual structural and visual review. Benchmark the exact committed workflow on more than one subject before assigning it a production batch.