Comfy-Org/ComfyUI · error · ValueError

SeedVR2TemporalMerge: chunk {i} has {chunk.shape[2]} latent

Error message

SeedVR2TemporalMerge: chunk {i} has {chunk.shape[2]} latent frames but chunk 0 has {first.shape[2]}; only the final chunk may be shorter.

What it means

In SeedVR2TemporalMerge, every chunk except the last must have exactly the same number of latent frames T as chunk 0; only the final chunk may be shorter (tail of the video). The overlap blending between consecutive chunks assumes aligned chunk lengths, so a shorter middle chunk would misalign the overlap window.

Source

Thrown at comfy_extras/nodes_seedvr.py:558

        if temporal_overlap < 0:
            raise ValueError(
                f"SeedVR2TemporalMerge: temporal_overlap must be >= 0; got {temporal_overlap}."
            )
        chunks = [entry["samples"] for entry in latents]
        first = chunks[0]
        if first.ndim != 5:
            raise ValueError(
                f"SeedVR2TemporalMerge: expected 5-D video latents (B, C, T, H, W); "
                f"chunk 0 has shape {tuple(first.shape)}."
            )
        for i, chunk in enumerate(chunks[1:], start=1):
            if chunk.shape[:2] != first.shape[:2] or chunk.shape[3:] != first.shape[3:]:
                raise ValueError(
                    f"SeedVR2TemporalMerge: chunk {i} shape {tuple(chunk.shape)} does not "
                    f"match chunk 0 shape {tuple(first.shape)} outside the temporal axis."
                )
            if i < len(chunks) - 1 and chunk.shape[2] != first.shape[2]:
                raise ValueError(
                    f"SeedVR2TemporalMerge: chunk {i} has {chunk.shape[2]} latent frames but "
                    f"chunk 0 has {first.shape[2]}; only the final chunk may be shorter."
                )

        out = latents[0].copy()
        out.pop("noise_mask", None)

        if len(chunks) == 1:
            out["samples"] = first
            return io.NodeOutput(out)
        if temporal_overlap == 0:
            out["samples"] = torch.cat(chunks, dim=2)
            return io.NodeOutput(out)

        chunk_latent = first.shape[2]
        step = chunk_latent - min(temporal_overlap, chunk_latent - 1)
        t_total = step * (len(chunks) - 1) + chunks[-1].shape[2]
        b, c, _, h, w = first.shape

View on GitHub (pinned to 1c6d8d45b3)

Solutions

Regenerate the offending middle chunk with the same frame length as chunk 0
Reorder the list so the short chunk is last
Fix the upstream chunking script to only emit a partial chunk at the tail

Defensive patterns

Strategy: validation

Validate before calling

if any(c.shape[2] != chunks[0].shape[2] for c in chunks[1:-1]):
    # move any short chunk to the end or regenerate it
    chunks.sort(key=lambda c: c.shape[2] == chunks[0].shape[2], reverse=True)

Prevention

Chunk scripts should emit equal-length chunks plus at most one tail chunk
Keep chunk order stable when collecting latents from parallel jobs

When it happens

Trigger: Passing a chunk list where a middle chunk was generated with a different length parameter — e.g. chunk 0 with length 81 and chunk 2 with length 41 — or reordering chunks so a short tail chunk ends up in the middle.

Common situations: Chunked video restoration where chunk length settings were changed between runs; manually assembling the latents list in the wrong order; a chunking script that emits a partial chunk anywhere but the end.

Related errors

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/8b3b82b16823de7a. Report an issue: GitHub.