Sushi Kitchen
Uramaki
Uramaki

ComfyUI

uramaki.comfyui

A Docker Compose file for ComfyUI alone, generated from the catalog and checked by the safety inspector. No account needed.

Node-based interface for diffusion image and video generation, where sampling, conditioning and model loading are explicit graph nodes that can be saved, shared and re-executed.

ComfyUI exposes the diffusion pipeline itself as a graph: checkpoint loading, text encoding, latent creation, sampling, decoding and post-processing are separate nodes with visible connections, so a workflow can be inspected, rearranged and reasoned about rather than driven through fixed form fields. Only the parts of the graph affected by a change are recomputed, which makes iterating on one stage fast. Workflows serialise to JSON and are embedded in the metadata of generated images, so a result carries the exact graph that produced it. Custom nodes extend it to video, audio and non-diffusion models, and an API mode runs saved graphs programmatically.

You know it worked when

  • The interface loads and the default workflow appears on the canvas.
  • Loader nodes list the models present in the mounted directories.
  • Queueing the default workflow produces an image.
  • A saved workflow reloads from its JSON file with all connections intact.
  • Generated images are written to the mounted output directory.

Known sharp edges

  • A GPU with matching drivers and container runtime configuration is needed; without one, generation either fails or takes minutes per image.
  • Model files are not included and must be placed in the correct subdirectory per type, since a checkpoint in the wrong folder simply does not appear in any node.
  • Custom nodes are third-party code executed inside the container, and a workflow shared from elsewhere will not load until every node it references is installed.
  • Model checkpoints carry independent licences, several of which restrict commercial use regardless of the software's own terms.
image-generationdiffusionnode-graph