The AI motion transfer war is heating up! Just days after Steady Dancer and OneToAll, we now have SCAIL Pose. Is it the best one yet? In this video, I break down this groundbreaking 3D pose estimation model that finally solves the "paper cutout" look of traditional 2D poses. We'll test its single person fluidity and its killer feature: stable dual person interaction.
- - - - - - - - - - - - - - - - - - - - - - - - - -
✨ Register on RunningHub to get 1000 FREE RH Coins!
💰 Plus, claim 100 extra coins every single day just by logging in.
🚀 Try it Now:
1️⃣AI APP: SCAIL_two person version:
https://www.runninghub.ai/ai-detail/2000499991896522753/?inviteCode=rh-v1291
2️⃣AI APP: SCAIL_Supports one person version:
https://www.runninghub.ai/ai-detail/2000950798836264961/?inviteCode=rh-v1291
🔽Free download workflow and models/Loras 🔽
Downlod link for SCAIL (huggingface) https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/tree/main/SCAIL Download link for
SCAIL Online storage: https://pan.quark.cn/s/9e456728ddab
LoRA:
[https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lightx2v](https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Lightx2v) [https://huggingface.co/Kijai/WanVideo_comfy/blob/main/WanAnimate_relight_lora_fp16.safetensors](https://huggingface.co/Kijai/WanVideo_comfy/blob/main/WanAnimate_relight_lora_fp16.safetensors)
Text encoder:
[https://huggingface.co/Kijai/WanVideo_comfy/blob/main/umt5-xxl-enc-bf16.safetensors](https://huggingface.co/Kijai/WanVideo_comfy/blob/main/umt5-xxl-enc-bf16.safetensors) VAE: [https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan2_1_VAE_bf16.safetensors](https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan2_1_VAE_bf16.safetensors) - - - - - - - - - - - - - - - - - - - - - - - - - -
💡 Why SCAIL Pose? Unlike OpenPose or DWPose which are 2D, SCAIL uses NLF (Neural Lift Field) to understand depth. This means when characters turn around or limbs cross, the AI knows what is in front and what is behind, resulting in much more stable animations. If this guide helps you create better AI videos, please SMASH that LIKE button and SUBSCRIBE for more ComfyUI tutorials! #SCAILPose #ComfyUI #AIMotionTransfer #Vid2Vid #GenerativeAI #StableDiffusion #AIAnimation #RunningHub
Description
FAQ
Comments (15)
Hello. I understand the main principle of all this but where do you find the "sticks" animations used as references? Thank you
You may use any video you wish to reference, but the main subject should be clearly visible. It need not be a 'sticks' animation,Once you have uploaded the reference video, the skeleton recognition model will convert it into a 'sticks' animation. You may watch my YouTube videos, which contain explanations.
@BiuBoomFlow Ok. I'm surprised people don't save the animation files and share them on some database. Thank you!!
@GlowingGuardianGirl Openpose / controlnet is old tech we've used since the 1.5 days. There was a website back then (obviously single poses, ~2gb) but it closed down. There's not much point when extracting directly from videos gets better & relatively faster over time, and models can handle prompts well enough for a more natural pose description in typical use cases.
Hi. I'm trying to make this work on my pod comfy, everything is set correctly but it crashes on " Render NLF Poses" node. No errors, just the comfy restarts by itself. then it's cancelled. any idea ?
ps. I have the log. It basically goes wrong just right after 'warming up' is finished :
NLF model warmed up
free(): invalid size /opt/supervisor-scripts/comfyui.sh: line 44: 6032 Aborted LD_PRELOAD=libtcmalloc_minimal.so.4 python main.py ${COMFYUI_ARGS} 2>&1 Terminated
I haven't encountered this issue myself. I tried asking Gemini, but the response seemed rather complex. You might need to raise an issue with the author on GitHub.
@BiuBoomFlow I see, thank you ! The github for Scail ?
@hboxgames132 https://github.com/zai-org/SCAIL-Pose You may wish to try this link.
How much vram is needed to run this? I keep getting "out of vram" on a 16 gig card.
You may wish to check whether the shared memory has been allocated sufficiently, Or max out the parameters of the block,or try downloading the GGUF version.:https://huggingface.co/vantagewithai/SCAIL-Preview-GGUF/tree/main
I tested this with 24GB of graphics memory. If it still doesn't work, you might need to consider lowering the generated resolution to get it running first, then adjust the other parameters.
why is onnx not working? i have made the files in the right place but dont detect it
Would WAN Animate Loras works ?
Like breast bounce Lora for exemple? Breasts on the preview felt like bricks
I've experimented with character-specific LORA, which effectively stabilises facial expressions during significant movements at weights of 0.3–0.5. However, I haven't yet tested the effect-type LORA you mentioned. You might wish to try it—both WAN2.1 and WAN2.2 are compatible, though you'll need to adjust the weight values accordingly.
wan2.1 loras work fine. some wan2.2 loras work, but they often degrade quality even at low weights. i have lots of scail uploads on my profile using 2.1 bounce loras