SCAIL-2 bf16 Loop FullLength Motion Transfer
Transfers the motion of a driving video onto a reference character image
character-animation
full-length
image-to-video
motion-transfer
pose-transfer
scail-2
video-to-video
wan2.1
5
528
Nodes & Models
PrimitiveNode
Display Int (rgthree)
CheckpointLoaderSimple
sam3.1_multiplex_fp16.safetensors
CLIPVisionLoader
clip_vision_h.safetensors
INTConstant
KSamplerSelect
VAELoader
wan_2.1_vae.safetensors
UNETLoader
wan2.1_14B_SCAIL_2_fp16.safetensors
CLIPLoader
umt5_xxl_fp8_e4m3fn_scaled.safetensors
FancyTimerNode
LoadImage
CLIPTextEncode
Reroute
WanChunkFeedForward
CLIPVisionEncode
ImageResizeKJv2
SAM3_VideoTrack
LoraLoaderModelOnly
Wan21_I2V_14B_lightx2v_cfg_step_distill_lora_rank64-wan-2-2-image-to-video-0ar6Pfb4.safetensors
SimpleCalculatorKJ
ModelSamplingSD3
BasicScheduler
SCAIL2ColoredMask
easy compare
WanSCAILToVideo
PreviewImage
SamplerCustom
VAEDecode
easy forLoopStart
GetImageRangeFromBatch
ColorTransfer
BatchImagesNode
easy forLoopEnd
ComfySwitchNode
VHS_LoadVideo
VHS_VideoInfo
VHS_VideoCombine
Transfers the motion of a driving video onto a reference character image and renders the entire clip length, using the Wan 2.1 14B SCAIL-2 preview model in full bf16 on a single H100 80GB (no GGUF quantization, no offload).
Pipeline: SAM3 tracks the subject in both the driving video and the reference image → SCAIL2ColoredMask builds colored pose/reference masks → WanSCAILToVideo conditions generation. The first 81-frame window is sampled in one pass (euler, 6 steps, LightX2V cfg-step-distill LoRA @ 0.8, shift 5), then a for-loop generates sliding 76-frame windows with a 5-frame overlap fed back as previous_frames, color-matches each chunk (ColorTransfer reinhard_lab) and stitches them until the full driving video is covered. The result is muxed with the original audio at source fps (h264 mp4).
Usage: load the reference image + driving video, set positive/negative prompts, set output resolution on the "Resize Pose Video" node — window/loop math is computed automatically from the source frame count.
Models: Wan21-14B-SCAIL-preview bf16, umt5_xxl fp8, clip_vision_h, wan_2.1_vae, LightX2V I2V rank64 LoRA, sam3.1_multiplex_fp16.
Custom nodes: ComfyUI-KJNodes, ComfyUI_Swwan, ComfyUI-easy-use, ComfyUI-VideoHelperSuite, rgthree-comfy, crt-nodes.
Read more
_1783698095782.gif?width=1400&height=620&quality=80&resize=cover)
_1783698095782.gif?width=104&height=104&quality=80&resize=cover)


_1783028563354.gif?width=400&height=300&quality=80&resize=cover)



