Comfy-Org/ComfyUI · error · ValueError

MiniMax H3 supports batch size 1

Error message

MiniMax H3 supports batch size 1

What it means

Raised in MiniMax H3's _forward when the video latent batch dimension is not 1. The model packs video and audio latents into a single PackedLayout keyed by (text_len, latent_t, lat_h, lat_w, audio_t) and its layout/attention code assumes exactly one sequence, so batching is not implemented. Any batched prompt fails fast with this ValueError.

Source

Thrown at comfy/ldm/minimax/model.py:539

        out = comfy.patcher_extension.WrapperExecutor.new_class_executor(
            self._forward,
            self,
            comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options)
        ).execute(x, timestep, context, transformer_options, minimax_payload=minimax_payload, **kwargs)

        if scale != 1.0:
            # d/d(sigma_v) of the carried variable
            out[1] = ((1.0 - scale) * (audio_src * carry)
                      + (1.0 + (scale - 1.0) * sigma_a).to(out[1].dtype) * out[1])
        return out

    def _forward(self, x, timestep, context, transformer_options={}, minimax_payload=None, **kwargs):
        video_x, audio_x = x[0], x[1]
        orig_t, orig_h, orig_w = video_x.shape[2], video_x.shape[3], video_x.shape[4]
        video_x = comfy.ldm.common_dit.pad_to_patch_size(video_x, self.patch_size)
        if video_x.shape[0] != 1:
            raise ValueError("MiniMax H3 supports batch size 1")
        payload = minimax_payload or {}
        device = video_x.device
        dtype = context.dtype  # compute dtype

        latent_t, lat_h, lat_w = video_x.shape[2], video_x.shape[3], video_x.shape[4]
        audio_t = audio_x.shape[-1]
        text_len = context.shape[1]
        # extra_conds prebuilds the layout once per sampling run
        layout = payload.get("layout")
        if layout is None or layout.signature != (text_len, latent_t, lat_h, lat_w, audio_t):
            layout = PackedLayout(text_len, latent_t, lat_h, lat_w, audio_t,
                                  keyframes=payload.get("keyframes"),
                                  refs=payload.get("refs"))

        # model_base passes model_sampling.timestep(sigma) = sigma * 1000
        shift_v = float(transformer_options.get("minimax_h3_sigma_shift_video", self.sigma_shift_video))
        shift_a = float(transformer_options.get("minimax_h3_sigma_shift_audio", self.sigma_shift_audio))
        sigma_v = (timestep.flatten()[0] / 1000.0).float().clamp(min=1e-6)

View on GitHub (pinned to 1c6d8d45b3)

Solutions

1. Set batch size to 1 for MiniMax H3 runs
2. Loop over prompts sequentially instead of batching
3. Check the workflow for nodes between the sampler and the model that expand the latent batch dimension

Example fix

# before
video_latent = video_latent.repeat(2, 1, 1, 1, 1)  # batch of 2
out = model((video_latent, audio_latent), t, ctx)
# after
video_latent = video_latent[0:1]  # batch of 1
out = model((video_latent, audio_latent), t, ctx)

assert video_latent.shape[0] == 1, f'MiniMax H3 requires batch size 1, got {video_latent.shape[0]}'

When it happens

Trigger: Calling the MiniMax H3 diffusion model's forward with video_x.shape[0] != 1 — e.g. batch size 2+, or accidentally leaving a batch dimension of size >1 after splitting/merging latents in a workflow.

Common situations: Setting batch_size > 1 in a ComfyUI workflow that includes the MiniMax H3 node; batching prompts for throughput; a latent-broadcast operation silently expanding the batch dim before the model call.

Related errors

• SeedVR2 expected {name} to be 5-D native latent, got shape {
• {type(model).__name__} must implement map_context_window_to_
• Passing a list or tuple of seeds to BatchedBrownianTree requ
• ar_video sampler requires 5-D video latents [B,C,T,H,W], got
• MiniMax Music3 prompt has {prompt_tokens} tokens; maximum is

AI-assisted analysis of Comfy-Org/ComfyUI@1c6d8d45b3 (2026-08-14). Data as JSON: /api/errors/7ea0649247406b83. Report an issue: GitHub.