<h2 id="basic-comfyui-workflow-for-chroma-diffusion-wvea4p0mw">Basic ComfyUI workflow for <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Chroma">Chroma Diffusion</a></h2><p></p><h3 id="necessary-extension-put-into-&quot;comfyuicustom_nodes&quot;-folder:-hox2dc2y7"><u>Necessary extension</u>, put into "ComfyUI\custom_nodes" folder:</h3><p><a target="_blank" rel="ugc" href="https://github.com/lodestone-rock/ComfyUI_FluxMod">https://github.com/lodestone-rock/ComfyUI_FluxMod</a></p><p></p><h3 id="models-folder-same-as-for-flux:-ylnif2hwc">Models folder same as for Flux:</h3><p>"ComfyUI\models\unet"</p><p>"ComfyUI\models\clip"</p><p></p><h3 id="chroma-gguf-models-run-same-as-normal-ones-use-&quot;load-chroma-diffusion-model&quot;-and-&quot;padding-removal&quot;-custom-nodes-with-t5-text-encoder-only.-bbwjvwc7j">Chroma GGUF models run same as normal ones, use "Load Chroma Diffusion Model" and "Padding Removal" custom nodes with T5 text encoder only.</h3><p></p><h3 id="checkpoints-for-gpu's-with-high-vram:-f7iezpgi0">Checkpoints for GPU's with high VRAM<strong>:</strong></h3><p><a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Chroma/tree/main">https://huggingface.co/lodestones/Chroma/tree/main</a></p><p></p><h3 id="checkpoints-for-gpu's-with-medium-vram:-b9hcc7vsi">Checkpoints for GPU's with medium VRAM<strong>:</strong></h3><p><a target="_blank" rel="ugc" href="https://huggingface.co/silveroxides/Chroma-GGUF/tree/main">https://huggingface.co/silveroxides/Chroma-GGUF/tree/main</a></p><p></p><h3 id="t5-text-encoder-(e4m3fn-preferred-for-medium-vram):-zujblu1qq">T5 text encoder (e4m3fn preferred for medium VRAM):</h3><p><a target="_blank" rel="ugc" href="https://civarchive.com/models/497255?modelVersionId=568405">https://civarchive.com/models/497255?modelVersionId=568405</a></p><p></p><p></p><h2 id="suggested-parameters-to-use:-ii6rlo0yo"><u>Suggested parameters to use:</u></h2><p>I prefer <u>euler ancestral</u> with <u>beta scheduler</u> as it follows prompt better than just <u>euler beta</u> or other combinations as I experimented.</p><p>It produces a lot of dots look like trypophobia hell and noise clusters at image edges. So "ModelSamplingFlux" with "max_shift" = 1.5 "base_shift" = 0.5 improves quality pretty much.</p><p></p><h3 id="other-parameters-same-as-for-flux:-jhwqz3n9w">Other parameters same as for Flux:</h3><p>good results at 25 steps with CFG 4.0-5.0, faster at 20 steps</p>

Files with SHA256: 6fcaac40bc56b81dea82b7299eae996f1bf9a59297213be77d49a4101a2bb92d