MiniMax H3 Easy-Use Workflow
T2VA and I2VA (first + optional last frame). Organized in four stages — MODELS → PASS 1 → PASS 2 → ENHANCE. Fast Turbo hybrid first pass at 0.2 MP, then 2× latent refine, FILM to 48 fps. High quality without touching the heavy SR/sharpen nodes unless you want them.
Leave both image loaders bypassed for T2VA. Un-bypass FIRST FRAME for I2VA. Un-bypass both for first-to-last.
Nodes required
DiffusionModelLoaderKJ
ModelAttentionBackend
SolAttnPatch
MiniMaxH3SigmaShift
MiniMaxH3ChunkFeedForward
VAELoader
ClipProjLoader
LTX_lora_loader
MiniMaxH3ImageToVideo
ResolutionSelector
ComfyMathExpression / easy float
RandomNoise, BasicGuider, KSamplerSelect, BasicScheduler, SamplerCustomAdvanced
LTXVSeparateAVLatent, LTXVConcatAVLatent
MinimaxH3LatentUpscaler3D
MiniMaxH3ConditioningUpscale
VAEDecode, VAEDecodeAudio
VRAM_Debug
FrameInterpolationModelLoader, FrameInterpolate
RTXVideoSuperResolution (muted)
ImageSharpenKJ (muted)
CreateVideo, SaveVideo
ModelPreviewOverrideKJ
LoadImage
Models required
minimaxh3\10Eros_Max_h3_TURBO-hybrid_beta5_int8.safetensors
minimax_h3_video_vae_int8_convrot.safetensors
minimax_h3_audio_vae_fp32.safetensors
qwen3-vl-4b-heretic_nvfp4.safetensors
mmh3-4b-ClipProj-celeb-mlp.safetensors
minimax_h3_latent_upscaler_3d_bf16.safetensors
film_net_fp16.safetensors
taeh3.safetensors
If you have any questions beyond this description, let me know!
Description
fixed some parameters causing the first few frames of videos to be fuzzy, also completes in 1 less step which results in faster generation.
FAQ
Comments (12)
all updated,all installed ,but
# ComfyUI Error Report ## Error Details - Node ID: 4:27:32 - Node Type: SamplerCustomAdvanced - Exception Type: RuntimeError - Exception Message: RuntimeError: shape mismatch: value tensor of shape [405, 96] cannot be broadcast to indexing result of shape [1620, 96] ## Stack Trace ``` File "K:\ComfyUI-Easy-Install\ComfyUI-Easy-Install\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 "K:\ComfyUI-Easy-Install\ComfyUI-Easy-Install\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 "K:\ComfyUI-Easy-Install\ComfyUI-Easy-Install\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(
i will run it and update what needs to be updated, i noticed i also have a connection misplaced. i will upload soon for this and a new version for my other all in one workflow also. thank you for the feedback
i uploaded a fix, try the update, i tested it immediately prior to uploading and it completed a generation
@billtomjimjob328 Nothing has changed at all. Perhaps, in addition to the poster, you could take a screenshot of the workflow? That would make it clear if a node from a different package has somehow ended up in the mix, or if there’s a ComfyUI issue with node rendering—like string field parameters changing, which is something I see often.
@artemzlykin369 did you run the workflow as-is, with everything installed from the errors list?
i just uploaded a test generation that i generated after asking that. i think maybe you have to update comfyui
edit: its the image input and i am working on it
i have uploaded the new workflow with the conditioning upscale node applied fixing the I2VA tensor mismatch. you will have to download the node pack for it to work but i got the same error as you beforehand, and after i applied the node it generated fully.
@billtomjimjob328 thx,it`s work
@artemzlykin369 no problem!
hello, thank you for the effort.
1- please have a link to this "LTX_lora_loader", can't find it exactly.
2- I don't do Turbo Loras, So I can just load my models and clip normally from the first screen, right? and adjust steps accordingly?
it is part of PlagueKind-Nodes - https://github.com/PlagueKind/ComfyUI-PlagueKind-Nodes
yes the h3 eros max model that is part of this workflow already includes turbo within the model itself and the steps are set accordingly to the model's preferences. and if you use a non turbo model you can change the steps on the main screen to what you like
another workflow another treasure hunt
MiniMaxH3ConditioningUpscale node found here https://github.com/rockerBOO/h3-latent-upscaler