My workflow with Live preview and 8-steps Lora. You must update ComfyUI in Manager to "ComfyUI Nightly Version", cause some nodes are still not up. For extra quality disable the 8-steps Lora and increase steps from 15 (or 8) to 50 and CFG to 3. Play with any option available (samlper/scheduler/cfg/steps/lora strength etc) to see what works best for you. The workflow uses 15 steps instead of 8 (slightly better quality), but you could go back to 8 if you want it fast.A very good 5B model, but unfortunately not on par with Krea 2 (my Scandinavian girls usually turn out to be Asian). Model (in models\diffusion_models): <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/blob/main/diffusion_models/moss_image_sft_bf16.safetensors">moss_image_sft_bf16.safetensors</a> or <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/blob/main/diffusion_models/moss_image_sft_int8_convrot.safetensors">moss_image_sft_int8_convrot.safetensors</a> Text encoders (in models\text_encoders): <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/blob/main/text_encoders/qwen3.5_9b_int8_convrot.safetensors">qwen3.5_9b_int8_convrot.safetensors</a> or <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/blob/main/text_encoders/qwen3.5_9b_w4a8_gptq.safetensors">qwen3.5_9b_w4a8_gptq.safetensors</a> vae (in models\vae): <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image/blob/main/vae/flux2-vae.safetensors">flux2-vae.safetensors</a> lodestones 8-step Lora (in models\loras\moss): <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/blob/main/lora_latest.safetensors">lora.safetensors</a>

Files with SHA256: fd55f5e4e81106f3954b0a8e71a0b26dda2e025b184cc59057481e26987d2016