Hi everyone, I’m currently using the ICE SCAIL2 Motion Control workflow for image-to-video generation, and the biggest issue I’m having is the hair. The face, body, pose, and motion transfer can look pretty good, but the hair often makes the final video look obviously AI-generated. It tends to lose the natural look of the original image once the motion starts. The texture can look too artificial, overly processed, or unnaturally defined, and it doesn’t always behave like real hair during movement. I’m trying to get a more natural smartphone-video look, so I don’t want the hair to look overly sharp, glossy, or “perfect.” I want it to keep softer, realistic texture and natural movement. Has anyone using ICE SCAIL2 Motion Control had the same problem? If so, what helped the most — different settings, LoRAs, resolution, sampler/steps, reference preparation, or a different post-processing/upscale method? Any advice would be appreciated.

Hi, I need some advice for upscaling Krea 2 images. My workflow already produces great, realistic images before upscaling. I use my own LoRA, Krea 2 Realism V2 from CivitAI, and a Krea 2 Smartphone LoRA. The problem is that my current upscaler, FLUX.2 Klein 9B + Qwen3 8B, makes the images way too sharp. Can anyone recommend a better upscaler or settings that preserve a softer, natural smartphone-photo look instead of making the image look like it was shot with a professional camera?

Hello :-) Can I just ask wich model for Lora I need train in your WAN 2.2 Lora Based Faceswap?I thought I was using the right model for training "Wan-AI/Wan2.2-TI2V-5B-Diffusers" but nothing happen in comfyui. Render just same video. Thanks

Guys i see more and more AI models with super detailed tattoos and really good quality, how is it possible to have such details Tattoos without drifting, every time i train a lora or create a dataset the tattoos are drifting and the results are bad, and with the Dataset workflows its not possible.

This worfklow will let you generate multiple angles of your character using references images which you will then be able to use as a LoRA training dataset. This is an updated version of my prior workflow using the new Qwen Image 2.1 edit capabilities Model:https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/diffusion_models/qwen_image_2.1_int8_convrot.safetensors Text encoder: https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/text_encoders/qwen3vl_8b_int8_convrot.safetensors 8step lora; https://huggingface.co/PrunaAI/Pruna-Qwen-Image-2.1/resolve/main/p_qwen_image_2.1_8step_v0.1.safetensors BFS: https://huggingface.co/Alissonerdx/BFS-Best-Face-Swap/blob/main/bfs_head_v1.1_qwen_2.1.safetensors