Experimented a bit before releasing but this model is the best out there right now easily for motion control. I will require a WAN character lora for the best possible likeness retention but can work without. Inputs are base reference image and video reference for movement The workflow attached as the free version is the official comfyui template , didn't test it as I kinda hate subgraphs Models should be the same, example was made with my workflow at medium resolution! My version with infinite length is already up for subs in the video workflows section Model: https://huggingface.co/Comfy-Org/SCAIL-2/tree/main/diffusion_models Sam3.1 in checkpoints: https://huggingface.co/Comfy-Org/sam3.1/blob/main/checkpoints/sam3.1_multiplex_fp16.safetensors Vae & Text encoders - wan_2.1_vae.safetensors: https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors - umt5_xxl_fp8_e4m3fn_scaled.safetensors: https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors - clip_vision_h.safetensors: https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/clip_vision/clip_vision_h.safetensors Loras - lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors: https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors

Going over this new model and my workflow and then jumping into lora training using runpod and AI-toolkit! The batch reference workflow in this video is available to subs and uses a custom node I made calling LM studio models to generate the json prompt based on reference images. Free version is simple one prompt reference workflow, you will need LM studio installed and setup to make it work properly, I can make a video on this if many people have difficulty setting it up This reference pipeline is not needed to use the model to its full potential but does speed up things 2 versions of this video, the sped up youtube one (I have a 15 mins limit), and the full 20 mins raw file. I will answers any questions regarding this model in this comment section

This workflow will let you recreate ANY image (nsf too) you want with your character as the main star. This can be used with a single image or with a lora. Obviously using a LoRa trained on Flux Klein will yield better results. Make sure you sub to get these workflows weeks in advance!!!!!!The best part is that you can change the image's feature only with prompting, no complex inpainting needed.I can't show the nsf results of this workflow on here but the results have been showcased in the discord for those who want to see. Enjoy! Model: https://huggingface.co/black-forest-labs/FLUX.2-klein-9b-fp8/blob/main/flux-2-klein-9b-fp8.safetensors Non gated Klein: https://modelscope.cn/models/black-forest-labs/FLUX.2-klein-9b-fp8/files Text encoder: https://huggingface.co/Comfy-Org/flux2-klein-9B/resolve/main/split_files/text_encoders/qwen_3_8b_fp8mixed.safetensors vae: https://huggingface.co/Comfy-Org/flux2-klein-9B/resolve/main/split_files/vae/flux2-vae.safetensors

https://x.com/icekiubAI/status/2045116980206702691?s=20 A WAN 2.2 low noise character Lora is required for this to work but I think it's one of the better solutions out there right now and I quickly go over the whole process in the video This is mostly a faceswap pipeline, hair color, body and clothes will be from the original video, select your reference videos with that in mind. We use WAN 2.2 T2V not I2V make sure you get that right Workflow in the zip My character on the left in the samples The video for this will be coming today or tomorrow WAN 2.2 T2V: https://huggingface.co/icekiub/WAN-2.2-T2V-FP8-NON-SCALED/resolve/main/WAN2.2t2vLOWNOISEFP8.safetensors 4step: https://huggingface.co/lightx2v/Wan2.2-Lightning/resolve/main/Wan2.2-T2V-A14B-4steps-lora-250928/low_noise_model.safetensors text encoder: https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/text_encoders/umt5_xxl_fp8_e4m3fn_scaled.safetensors vae: https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/blob/main/split_files/vae/wan_2.1_vae.safetensors

Covers the whole thing Prodigy with a learning rate of 1 can be used instead of AdamW8bit