Moe Fussion Flux/Chroma EXPERIMENTAL DO NOT USE UNLESS YOU WANT TO EXPERIMENT WITH FLUX AND KNOW WHAT YOU ARE DOING! Moe Fussion focuses on providing wide spectrum of models suitable for different purposes. This model is under Apache 2.0 license. Please read a rest to know how to use the model. General Info: This is highly experimental model which bases on Chroma (Flux). You can support Lodestone <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Chroma">https://huggingface.co/lodestones/Chroma</a> in theirs efforts of creating an edited Flux architecture while as well adding NSFW concepts inside. As I said, this is highly experimental and Chroma itself is not finished either, so don't expect things to go smooth. But if you want Flux anime NSFW, this is the way to go, at least for now. Chroma is mainly furry oriented model so, what I did is shifting that to more anime and adding default style. You can either use it with natural language or tags but I was testing on tags, you can try prompting for artists but that will be limited likely. Chroma is full scale finetune, this mine here is just LoRA scale on top of it with some merging. Also since it's custom architecture you need to use custom ComfyUI workflow (in downloads, the zip file - inside is a JSON) and install this extension to comfy as well <a target="_blank" rel="ugc" href="https://github.com/lodestone-rock/ComfyUI_FluxMod">https://github.com/lodestone-rock/ComfyUI_FluxMod</a> I can basically guarantee you no other app or online service will be able to run this at this point, that will likely change as Chroma will probably become more popular but not yet. You need 3 components to download, this (place it in unet folder in comfy models), T5 Encoder (if you already have is fine otherwise here <a target="_blank" rel="ugc" href="https://huggingface.co/city96/t5-v1_1-xxl-encoder-bf16/blob/main/model.safetensors">https://huggingface.co/city96/t5-v1_1-xxl-encoder-bf16/blob/main/model.safetensors</a> rename however you want and place to clip folder), and AE (<a target="_blank" rel="ugc" href="https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/ae.safetensors">https://huggingface.co/black-forest-labs/FLUX.1-schnell/blob/main/ae.safetensors</a> place to vae folder). If you are missing something, your best bet is to train your own LoRA since well, there are none, neither LoRAs nur other finetunes of this. I made an custom code for LoRA training so if you want, here: <a target="_blank" rel="ugc" href="https://github.com/jilek-josef/flow-chroma">https://github.com/jilek-josef/flow-chroma</a> As for other stuff like best configuration or prompts or even negative prompt... well, you have to experiment, there are none. A working workflow with defaults is attached however.

Files with SHA256: da53ceab28e9b9b01f6a2ff2df12769527bca373c8ab6cfaafa949d760117ea0