nodes/olivv-cs-ComfyUI-FunPack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix crash: i2v anchor must no-op on packed AV (LTXAV) denoise masks

Browse Source 
The i2v anchor (DynamicConditioning port) crashed on LTXAV with 'not enough values
to unpack (expected 5, got 3)': comfy's process_timestep patchifies the video denoise
mask expecting unpacked 5D [B,C,F,H,W], but LTXAV packs the AV latent so its mask is
3D [B,C,N]. The else-branch (first_frame_only=False) modified + propagated the 3D mask
unguarded -> patchify blew up.

Hard-guard the whole denoise_mask_fn to genuine 5D masks; packed/non-5D -> clean no-op
(one-time log). The technique is fundamentally unpacked-video-only, so on LTXAV it does
nothing -- documented in the tooltip. Native packed-safe carry-strength is the follow-up.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
digital-garbage
2026-06-05 20:54:49 +03:00
co-authored by Claude Opus 4.8
parent a5a566b0d0
commit 04d00fcecd
1 changed files with 12 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
samplers.py
+12 -3
View File
@@ -2072,7 +2072,7 @@ class FunPackLTXAVSceneChainSampler:
}),
"i2v_anchor_power": ("FLOAT", {
"default": 1.0, "min": 1.0, "max": 2.0, "step": 0.01,
"tooltip": "Progressively strengthen i2v adherence to the conditioning frame as denoising proceeds (port of LTXV DynamicConditioning): the denoise mask is raised to power**step. 1.0 = off (no change). ~1.3 = noticeable anti-drift. EXPERIMENTAL on packed AV latents — validate in ComfyUI. ~zero overhead.",
"tooltip": "Progressively strengthen i2v adherence to the conditioning frame as denoising proceeds (port of LTXV DynamicConditioning): the denoise mask is raised to power**step. 1.0 = off. NOTE: only works on UNPACKED 5D video masks; on PACKED audio+video (LTXAV) it safely no-ops (the mask is 3D and comfy's per-token-timestep patchify needs 5D). ~zero overhead.",
}),
"i2v_anchor_first_frame_only": ("BOOLEAN", {
"default": True,
@@ -2811,10 +2811,19 @@ class FunPackLTXAVSceneChainSampler:
exp = p ** _find_step(sigma, step_sigmas)
if abs(exp - 1.0) < 1e-6:
return denoise_mask
# This technique (LTXV DynamicConditioning) assumes an UNPACKED 5D video mask
# [B,C,F,H,W]: comfy's LTXAV process_timestep patchifies the mask expecting 5D.
# LTXAV PACKS the AV latent, so its denoise mask is 3D [B,C,N] — touching or
# propagating that crashes patchify (expected 5, got 3). So operate ONLY on genuine
# 5D masks; on packed AV (or anything else) this is a clean no-op.
if denoise_mask.dim() != 5:
if not getattr(self, "_i2v_anchor_packed_warned", False):
self._i2v_anchor_packed_warned = True
print("[FunPackSceneChain] i2v anchor: packed/non-5D denoise mask "
f"(ndim={denoise_mask.dim()}) -> no-op (technique needs unpacked video).")
return denoise_mask
mask = denoise_mask.clone()
if first_frame_only:
if mask.dim() < 3 or mask.shape[2] < 1:
return denoise_mask # not a [B,C,F,...] video mask -> don't touch (e.g. packed)
try:
num_channels = inner.model_patcher.model.diffusion_model.in_channels
except Exception: