Nodes/comfyui-scail2-infinity/SCAIL-2 Infinity (auto window)
ComfyUI Node

SCAIL-2 Infinity (auto window)

All-in-one SCAIL-2 long-video node. Internally loops 81-frame chunks (with a 5-frame overlap) over the driving pose video, stitching them into one continuous video. Replaces the WanSCAILToVideo + KSampler + VAEDecode + manual chunk-wiring graph.

By collbroGTR·Created 2 months ago·Updated 2 months ago· 42
SCAIL-2 Infinity (auto window)
  • positive
  • negative
  • model
  • vae
  • pose_video
  • pose_video_mask
  • reference_image
  • reference_image_mask
  • clip_vision_output
  • images
  • latent
  • total_frames
width512
height896
seed0
steps20
cfg6.0
sampler_name
scheduler
denoise1.00
window_length81
previous_frame_count5
max_frames0
decode_tiledfalse
vary_seed_per_windowfalse
pose_strength1.00
pose_start0.00
pose_end1.00
replacement_modefalse
Categoryvideo/scail2

Inputs (26)

NameTypeDefaultDescription
positiveCONDITIONING
negativeCONDITIONING
modelMODEL
vaeVAE
widthINT51232–16384
heightINT89632–16384
seedINT00–18446744073709550000
stepsINT201–10000
cfgFLOAT6.00–100
sampler_nameCOMBO44 options: euler, euler_cfg_pp, euler_ancestral, euler_ancestral_cfg_pp, heun, heunpp2, +38
schedulerCOMBO9 options: simple, sgm_uniform, karras, exponential, ddim_uniform, beta, +3
denoiseFLOAT1.000–1
window_lengthINT815–16384Frames per chunk. SCAIL-2 was trained at 81; changing this is not recommended.
previous_frame_countINT51–16384Overlap frames anchored from the previous chunk. SCAIL-2 trained at 5.
max_framesINT00–16384Hard cap on total output frames. 0 = run until the driving pose video is exhausted.
decode_tiledBOOLEANfalseUse tiled VAE decode to bound decode VRAM at high resolution.
vary_seed_per_windowBOOLEANfalseAdd the window index to the seed each chunk. Off = same seed every chunk (anchored continuity).
pose_strengthFLOAT1.000–10
pose_startFLOAT0.000–1
pose_endFLOAT1.000–1
pose_videooptIMAGEDriving pose video. Its length determines how many frames are generated. Downscaled to half resolution internally.
pose_video_maskoptIMAGESCAIL-2 colored per-identity SAM3 mask video, same resolution as pose_video.
replacement_modeoptBOOLEANfalseFalse = Animation Mode (black-bg mask). True = Replacement Mode (white-bg mask).
reference_imageoptIMAGEReference character image. For multiple references composite them on a single image.
reference_image_maskoptIMAGESCAIL-2 colored reference mask, same resolution as reference_image.
clip_vision_outputoptCLIP_VISION_OUTPUTCLIP vision features for conditioning.

Outputs (3)

NameTypeDescription
imagesIMAGEFull stitched video.
latentLATENTConcatenated, overlap-free latent of the full video.
total_framesINT