<h3 id="comfyui-workflow">ComfyUI Workflow</h3><ul><li>Requires: <a target="_blank" rel="ugc" href="https://github.com/r-vage/ComfyUI_Eclipse">ComfyUI_Eclipse 3.4.x</a></li></ul>its a very basic workflow that uses nodes from Eclipse which supports 8 Backends:<ul><li>transformers</li><li>gguf (llama-cpp-python)</li><li>ollama (docker), vllm (docker), sglang (docker), llama.cpp (docker)</li><li>wd14 (onnx)</li><li>yolo</li></ul><ul><li>Image Description (VLM / WD14) or text input (LLM)</li><li>Detection (Face, Eyes, Hands, and other areas ;) using Florence2 and Yolo (qwen detections are usally meh... but works very good with huihui qwen 3.5 9b claude 4.6</li></ul>try claude for image to prompt descriptions / detection (also very good):<ul><li>In most cases, the model doesn't mince words and describes exactly what it "sees".</li><li>ollama: huihui_ai/qwen3.5-abliterated:9b-Claude</li><li>transformers: <a target="_blank" rel="ugc" href="https://huggingface.co/huihui-ai/Huihui-Qwen3.5-9B-Claude-4.6-Opus-abliterated">https://huggingface.co/huihui-ai/Huihui-Qwen3.5-9B-Claude-4.6-Opus-abliterated</a></li></ul>

Image to Prompt & Detection

<ul><li>The latest Smart LM / Smart Detection update requires reloading the nodes in existing workflows — I removed the compat remapping because it was getting too messy to maintain.</li><li>The widget order is also reorganized now so it makes sense by itself, instead of being reshuffled in JS every time a new widget is added (I'm not planning to add more… but just in case).</li><li>After downloading the AceStep workflow I noticed a big gap in the settings that Ollama and the other Docker backends actually support — some of them were even silently dropped 😒. Since I don't want to use the Ollama server directly but go through the Docker backends the LM / Detection nodes already use, Eclipse 3.4 ships a lot of adjustments and fixes there. But as mentioned, you'll have to reload the nodes in older workflows. Sorry about that.</li><li>If you haven't installed Docker yet, I'd suggest taking the time to set it up — inference is much faster than Transformers, and the Ollama registry has a lot of models now (both standard and abliterated).</li><li>to get tags in the direct chat example enable multi task and set task 2 to natural lang to tags ;)</li></ul>

Files with SHA256: 16321864986eb6ae2da2b0e2b8c8d3adcf9951c10ca5d95a4370142f10651cf9