comfyui-ollama-nodes
Add LLM workflows, including image recognition (vision), to ComfyUI via Ollama
Currently, it assumes that an Ollama server is running at http://127.0.0.1:11434 and accessible by the ComfyUI backend.
Screenshots
Pulling models from the Ollama Model Libary with download progress bar:
Generating text descriptions of loaded images:
TODO:
[X] Implement model pulling node
[x] Implement UI progress bar updates when pulling with stream=True
[x] Impelement Huggingface Hub model downloader node
[ ] Implement progress bar updates when download models
[x] ~~Implement model loading node~~ Theyre loaded dynamically on generate, it’s good to unload them from GPU to leave room for image generation models in the same workflow.
[x] Implement generate node
[ ] Impelment generate node with token streaming (I think it’s just a UI limitation with ShowText class)
[x] Implement generate node with vision model (can take image batch as input!)
[ ] Implement chat node (likely requires new frontent node development)
[ ] Implement model converter node (saftetensor to GGUF)
[ ] Implement quantization node
[ ] Test compatability with SaltAI LLM tools (LlamaIndex)
Similar Nodes
The following node packs are similar and effort will be made to integrate seemlessly with them:
https://github.com/daniel-lewis-ab/ComfyUI-Llama
https://github.com/get-salt-AI/SaltAI_Language_Toolkit
https://github.com/alisson-anjos/ComfyUI-Ollama-Describer
Development
If you’d like to contribute, please open a Git Issue describing what you’d like to contribute. See ComfyOrg docs for instructions on getting started developing custom nodes.
Attributions
logger.py taken from https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite, GPL-3.0 license
Ollama MIT license, based on llama.cpp MIT license
huggingface_hub python library Apache-2.0 license