Shape mismatch in MiniMax H3 audio conditioning when combining video continuation ("OG Video Has Audio") with a standalone audio-timbre reference

When generating a MiniMax H3 video continuation using a source video with its own audio (continuation guide), combined with a separate standalone audio reference clip used for voice-timbre conditioning, generation fails with a tensor shape mismatch. Removing the standalone audio reference (keeping only the continuation) succeeds; removing the continuation (keeping only the standalone reference) also succeeds. The failure only occurs when both are present together.

Steps to reproduce:

Set up a MiniMax H3 Reference-to-Video generation with a continuation source: feed in an existing generated video (with its own audio track) as the continuation base, using a short overlap window (~24 frames / 1 second in my case).
Enable audio carry-over from that continuation source ("include this clip's audio" / equivalent toggle).
Additionally connect a separate, standalone audio clip as a voice-timbre reference ( tag), independent of the continuation video.
Queue the prompt.

Expected behavior: Generation proceeds normally, using the continuation audio for temporal carry-over and the standalone reference for voice timbre.

Actual behavior: Crashes during sampling with:

RuntimeError: shape mismatch: value tensor of shape [938, 32] cannot be broadcast to indexing result of shape [1010, 32]

Full traceback bottoms out at:

File "comfy/ldm/minimax/model.py", line 711, in _forward
all_audio_rows[~audio_update] = cond_audio_rows
RuntimeError: shape mismatch: value tensor of shape [938, 32] cannot be broadcast to indexing result of shape [1010, 32]

Analysis: audio_update appears to mark which audio-row positions are supplied by the continuation source; ~audio_update should then be filled from cond_audio_rows derived from the standalone audio reference. The row count encoded from the standalone reference (938) doesn't match the count of positions the mask expects to be filled (1010) — these two lengths appear to be computed independently and don't reconcile when both audio sources are present together.

This looks related to the class of bugs discussed in #15439 (where combining guides/keyframes with references caused PackedLayout to emit rows with mismatched/absent backing latents), but reproduces on current master (commit c75d8c966c29cb0392259af791f43373315b72db, 2026-09-12) after that PR merged, so appears to be a distinct, still-open case specific to the audio-only combination rather than the image/video-reference case that PR addressed.

Environment:

ComfyUI commit: c75d8c966c29cb0392259af791f43373315b72db (2026-09-12), branch master
Windows portable install (NVIDIA)
Relevant custom nodes present but not implicated in the crash itself (traceback bottoms out in core comfy/ldm/minimax/model.py)

**Full traceback:**

[INFO] Comfy model compiler graph breaks: 0, rogues: 10
[ERROR] !!! Exception during processing !!! shape mismatch: value tensor of shape [938, 32] cannot be broadcast to indexing result of shape [1010, 32]
[ERROR] Traceback (most recent call last):
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 545, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-dapaoAPI\node_execution_gate.py", line 125, in get_output_data_waiting
result = await current(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 344, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 177, in async_map_node_over_list_with_metadata
results = await original_map_node_over_list(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
...<2 lines>...
)
^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 318, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\execution.py", line 306, in process_inputs
result = f(**inputs)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func
return method(locked_class, **inputs)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy_api\latest\_io.py", line 2046, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy_extras\nodes_custom_sampler.py", line 1055, in execute
samples = guider.sample(noise.generate_noise(latent), latent_image, sampler, sigmas, denoise_mask=noise_mask, callback=callback, disable_pbar=disable_pbar, seed=noise.seed)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1335, in sample
output = executor.execute(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1262, in outer_sample
output = self.inner_sample(noise, latent_image, device, sampler, sigmas, denoise_mask, callback, disable_pbar, seed, latent_shapes=latent_shapes)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1237, in inner_sample
samples = executor.execute(self, sigmas, extra_args, callback, noise, latent_image, denoise_mask, disable_pbar)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-yaser-nodes\nodes\tiling_utils.py", line 34, in KSAMPLER_sample
return orig_fn(*args, **kwargs)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1005, in sample
samples = self.sampler_function(model_k, noise, sigmas, extra_args=extra_args, callback=k_callback, disable=disable_pbar, **self.extra_options)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\utils\_contextlib.py", line 124, in decorate_context
return func(*args, **kwargs)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\k_diffusion\sampling.py", line 1619, in sample_er_sde
denoised = model(x, sigmas[i] * s_in, **extra_args)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 640, in __call__
out = self.inner_model(x, sigma, model_options=model_options, seed=seed)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1208, in __call__
return self.outer_predict_noise(*args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1215, in outer_predict_noise
).execute(x, timestep, model_options, seed)
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 1218, in predict_noise
return sampling_function(self.inner_model, x, timestep, self.conds.get("negative", None), self.conds.get("positive", None), self.cfg, model_options=model_options, seed=seed)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 620, in sampling_function
out = calc_cond_batch(model, conds, x, timestep, model_options)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 211, in calc_cond_batch
return _calc_cond_batch_outer(model, conds, x_in, timestep, model_options)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 219, in _calc_cond_batch_outer
return executor.execute(model, conds, x_in, timestep, model_options)
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\samplers.py", line 335, in _calc_cond_batch
output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\model_base.py", line 209, in apply_model
return comfy.patcher_extension.WrapperExecutor.new_class_executor(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
...<2 lines>...
comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.APPLY_MODEL, transformer_options)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
).execute(x, t, c_concat, c_crossattn, control, transformer_options, **kwargs)
~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\model_base.py", line 253, in _apply_model
model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1778, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\nn\modules\module.py", line 1789, in _call_impl
return forward_call(*args, **kwargs)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\ldm\minimax\model.py", line 577, in forward
graph_out = comfy.patcher_extension.WrapperExecutor.new_class_executor(
...<2 lines>...
comfy.patcher_extension.get_all_wrappers(comfy.patcher_extension.WrappersMP.DIFFUSION_MODEL, transformer_options)
).execute(x, timestep, context, transformer_options, minimax_payload=minimax_payload,
denoise_mask=denoise_mask, audio_denoise_mask=audio_denoise_mask, **kwargs)
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\patcher_extension.py", line 113, in execute
return self.original(*args, **kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "G:\Users\Laszlo\Documents\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\comfy\ldm\minimax\model.py", line 711, in _forward
all_audio_rows[~audio_update] = cond_audio_rows
~~~~~~~~~~~~~~^^^^^^^^^^^^^^^
RuntimeError: shape mismatch: value tensor of shape [938, 32] cannot be broadcast to indexing result of shape [1010, 32]

Start in comfy/ldm/minimax/model.py at _forward around line 711, then reproduce the MiniMax H3 continuation-plus-standalone-audio workflow described in the issue. Trace how audio_update and cond_audio_rows are sized for both sources; done means the combined case no longer raises the reported shape mismatch and generation proceeds.

Written by the indexing model from the issue text.