Nodes/ComfyUI-NynxzH3/H3 Decode Frames
ComfyUI Node

H3 Decode Frames

Decodes chosen frames of an H3 video latent without decoding the clip. Exact — it reproduces the VAE's temporal chunking and cross-fade — and costs one chunk per frame, so pulling the last frame of a 124-frame clip is ~1/7th of a full decode. The point of it is using a video model as an edit model.

By Nynxz·Created 14 days ago·Updated 5 days ago· 4
H3 Decode Frames
  • samples
  • vae
  • IMAGE
  • total_frames
index-1
count1
stride1
CategoryNynxz/H3/Latent

Inputs (5)

NameTypeDefaultDescription
samplesLATENTAn H3 latent. The audio stream of an AV latent is ignored.
vaeVAEThe MiniMax H3 video VAE.
indexINT-1-4096–4096Pixel frame to decode. Negative counts from the end, so -1 is the last frame and 0 the first — the two an edit workflow usually wants.
countINT11–256How many frames to take. Each costs its own chunk decode unless it shares one with a neighbour, so consecutive frames are much cheaper than spread-out ones.
strideINT11–1024Gap between them. count=6 stride=20 samples a clip at six points; count=8 stride=1 gives eight consecutive frames out of one or two chunks.

Outputs (2)

NameTypeDescription
IMAGEIMAGE
total_framesINTFrames a full decode of this latent would produce.