Load Random Image

Cut frames from the start of a WAN video latent before VAE decode. The count is in pixel frames, converted to latent frames (4x temporal compression) and…

Nodes for loading images, writing prompts and video/music work. Image: random/sequential loader for any folder, touch-friendly gallery picker for the input folder, loader with an on-node crop editor, booru tag search (Gelbooru, yande.re, Konachan, Safebooru, Xbooru), Ollama anatomy-defect filter. Text: 52prompts.com generators, MiniMax H3 video prompt writer (Ollama or a Qwen/Gemma CLIP, validated, with a Generate button). Latent: WAN latent start trim. Audio: YuE2 song generation. 13 example workflows.

Bart's ComfyUI Nodes

samples

trim_start_frames

Cut frames from the start of a WAN video latent before VAE decode. The count is in pixel frames, converted to latent frames (4x temporal compression) and snapped to the nearest latent boundary.

Name	Type	Default	Description
samples	LATENT		—
trim_start_frames	INT	00–16384	Pixel frames to remove from the start. Converted to latent frames as N / 4 rounded to the nearest whole frame, halves up (not Python's round-half-even); multiples of 4 are exact.

Name	Type	Description
samples	LATENT	—
removed_frames	INT	—
frames_left	INT	—

Bart's WAN Latent Trim Start 🎞️

Inputs (2)

Outputs (3)