Generate latent videos with audio, continue scenes, and assemble clips, upscale with MiniMax H3 in ComfyUI.
Results are saved as .mmh3 files. Along with the media, the file stores settings, references, and the internal generation state - latents. This lets you return to a project later and continue or upscale the scene.
Features:
Automatic latent stitching, context and motion inheritance
Latent upscaling
In development:
Lip sync
ControlNet
Inpainting
Audio
Installation
Download the repository https://github.com/einhorn13/mmh3_media via Code → Download ZIP and extract it to
ComfyUI/custom_nodes/ComfyUI_mmh3_media. The__init__.pyfile must be located directly inside this folder, without an extra nested repository directory.Place the models in the folders listed below and select them in the workflow loaders.
Restart ComfyUI, refresh the browser, and open a JSON file from workflows.
Description
Comments (6)
optimization node bugged.
"R:\ComfyUI_windows_portable\ComfyUI\custom_nodes\mmh3_media-main.mmh3_media.errors.MMH3ResourceError: Missing selected optimization node(s): PatchSageAttentionKJ"
I HAVE sage attention and its working in other workflows.
Thanks for giving it a try and sharing the bug! I fixed the Sage Attention naming issue, so it should be working now. Just update nodes.
Also critical lack of pipelines descriptions. Your pack is too cryptic right now, just impossible to understand how use references at all for example. You changed way of references prompting but not give any info how prompt references now. With basic minimax syntax (subject_definitions: and retention_analysis:) references now ignored.
I'm on my way to make workflows simple and clear. stay tuned.
ref2v contiunuation throws error.
[ERROR] !!! Exception during processing !!! [WinError 5] Отказано в доступе: 'R:\\ComfyUI_windows_portable\\ComfyUI\\temp\\mmh3_media_cache\\session-1740-b97a7d8e76\\c6f28d1a1638301cf4ef357a3bb7ec1296451052742f7f0d174cdcf3132a1d36.safetensors'
[ERROR] Traceback (most recent call last):
File "R:\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 "R:\ComfyUI_windows_portable\ComfyUI\execution.py", line 344, in get_output_data
return_values = await asyncmap_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 "R:\ComfyUI_windows_portable\ComfyUI\execution.py", line 318, in asyncmap_node_over_list
await process_inputs(input_dict, i)
File "R:\ComfyUI_windows_portable\ComfyUI\execution.py", line 306, in process_inputs
result = f(**inputs)
File "R:\ComfyUI_windows_portable\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func
return method(locked_class, **inputs)
File "R:\ComfyUI_windows_portable\ComfyUI\comfy_api\latest\_io.py", line 1990, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
File "R:\ComfyUI_windows_portable\ComfyUI\custom_nodes\mmh3_media-main\mmh3_media\nodes_h3.py", line 258, in execute
source_latent = get_resource_payload(packet, resource)
File "R:\ComfyUI_windows_portable\ComfyUI\custom_nodes\mmh3_media-main\mmh3_media\archive.py", line 303, in get_resource_payload
path.unlink(missing_ok=True)
~~~~~~~~~~~^^^^^^^^^^^^^^^^^
File "pathlib\_local.py", line 746, in unlink
PermissionError: [WinError 5] Отказано в доступе: 'R:\\ComfyUI_windows_portable\\ComfyUI\\temp\\mmh3_media_cache\\session-1740-b97a7d8e76\\c6f28d1a1638301cf4ef357a3bb7ec1296451052742f7f0d174cdcf3132a1d36.safetensors'
Thank you for your feedback. I already fixed this issue yesterday. Could you please update the nodes and the workflow from github repo and check whether you're still getting the error?