digital-garbageandClaude Opus 5.5
6838e3b997
Cuts decoded frames and the same time off the sound. In both H3 presets, 0 by default. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
27 lines
1.0 KiB
Python
27 lines
1.0 KiB
Python
"""Cut the opening: let the starting image do its job, then remove it from the clip.
|
|
|
|
An i2v anchor pinned at full strength transfers identity, style and composition
|
|
better than anything that softens it (ALG blurs it, identity tokens approximate
|
|
it) -- but it is also literally the first frame you see, so every i2v shot opens
|
|
on the exact reference still. Generate normally, then drop N frames off the
|
|
front, and the sound by the same amount of time.
|
|
|
|
Cut on DECODED frames, never on the latent (v4 learned both reasons): a causal
|
|
video VAE decodes latent frame 0 as the time origin, so a promoted survivor
|
|
decodes wrong; and H3's latent sits on a 5k+2 grid, so most cut counts leave a
|
|
length the VAE cannot decode at all. On pixels, N means N.
|
|
|
|
Nothing is regrown to replace what is cut -- v4 tried, and the invented ending
|
|
consistently moved worse. The clip just comes out shorter.
|
|
"""
|
|
|
|
from .nodes import FunPackCutOpening
|
|
|
|
ID = "output_cut_opening"
|
|
TITLE = "Cut opening"
|
|
STAGE = "post"
|
|
CATEGORY = "post"
|
|
STATUS = "proven"
|
|
|
|
NODES = [FunPackCutOpening]
|