Set up ComfyUI

Install the tested composition stack

The committed composition and post-processing benchmarks use three public model files:

FileComfyUI folderSHA-256License named by the model card
sd_xl_base_1.0.safetensorsmodels/checkpoints31e35c80fc4829d14f90153f4c74cd59c90b779f6afe05a74cd6120b893f7e5bCreativeML Open RAIL++-M
pixel-art-xl.safetensorsmodels/loras4234637cb80c998f41e348e6a6cb6bc20d8d038b2b0f256b6129b3b5e353eef7CreativeML OpenRAIL-M
birefnet.safetensorsmodels/background_removal9ab37426bf4de0567af6b5d21b16151357149139362e6e8992021b8ce356a154MIT

Download each file into the named folder, verify its checksum, then confirm the checkpoint, LoRA, and background-removal model appear in ComfyUI. The files are about 6.9 GB, 171 MB, and 444 MB. They are not bundled with PixelKiln. Read the model cards before distributing the models or their outputs.

The environment benchmark renders at 1024×1024, where SDXL has enough room to compose the scene, then uses ComfyUI's core ImageScale node with nearest-exact to write the requested 256px or 384px PNG. Asset width and height are therefore bound to the scale node, not the latent node. Those files prove exact delivery dimensions; they do not prove a native pixel grid. The separate native-grid benchmark saves the model canvas and reconstructs its implied cells afterward.

You can reproduce the four samples with the committed ComfyUI benchmark project.