MiniMax-H3-Turbo-Lora-ComfyUI

MiniMax-H3 Turbo LoRAs for ComfyUI

This repository contains MiniMax-H3 Turbo and few-step LoRAs converted and optimized for ComfyUI.

These LoRAs accelerate MiniMax-H3 video and synchronized-audio generation by reducing the required number of sampling steps.

---

Update — HyperFlow 8-Step LoRAs

Four HyperFlow 8-step MiniMax-H3 LoRAs are now included.

They are derived from the original HyperFlow release by Video Rebirth:

• Video-Rebirth/hyperflow — GitHub
• videorebirth/hyperflow — Hugging Face

HyperFlow is an 8-step MiniMax-H3 LoRA designed for the MiniMax-H3 T2VA, FL2VA and Ref2VA workflows.

The upstream HyperFlow implementation uses its own fixed 8-step sigma schedule. For these ComfyUI conversions, the recommended manual sigma sequence is:

1.0, 0.9939097854, 0.9842874953, 0.9660637197, 0.9230769231, 0.8349423898, 0.6968520491, 0.4687533149, 0.0

This is a nine-point sigma sequence defining eight denoising intervals.

HyperFlow files

Use the non-pruned files with the corresponding full MiniMax-H3 base and the pruned files with the compatible pruned/curve-form base.

Recommended manual HyperFlow schedule

• Denoising steps: 8
• Sigma points: 9
• LoRA strength: start at 1.0
• Manual sigmas: 1.0, 0.9939097854, 0.9842874953, 0.9660637197, 0.9230769231, 0.8349423898, 0.6968520491, 0.4687533149, 0.0 OR euler / simple, normal

When using the original Video Rebirth implementation, follow the upstream HyperFlow scheduler and loader rather than replacing its built-in schedule manually.

---

Update — TaoMate 3-Step EMA and VDN-H3 8-Step Extracts (experimental)

TaoMate FL2VA 3-step EMA

Newly added LoRA, located in the experimental/ folder:

minimax_h3_taomate_fl2va_3step_ema_comfyui.safetensors

Manual recommended sigmas: 3-step 1.0, 0.961165, 0.853333, 0.0 <br> 4-step 1.0, 0.970874, 0.907249, 0.640000, 0.0

VDN-H3 8-step extracts

Three LoRAs extracted from VDN-H3 8 step:

• minimax_h3_dmd_ref2va_8step_turbo_pruned.safetensors
• minimax_h3_dmd_fl2va_8step_turbo_pruned.safetensors
• minimax_h3_dmd_8step_turbo.safetensors

The main 8-step LoRA works on both FL2VA and Ref2VA. If you are running a pruned base, choose the pruned version that corresponds to your base — the pruned versions need their own matching pruned base.

---

Update — LightX2V FL2V and Ref2V Dynamic-Rank LoRAs

Five dynamically resized BF16 LoRAs are now included. Their source weights were published in the correct MiniMax-H3 ComfyUI layout by:

lightx2v/Minimax-h3-Turbo

New files

The LightX2V files were already supplied in the correct ComfyUI namespace and QKV layout. No namespace conversion or AdaLN pruning was required.

How the LoRAs were resized

Each LoRA was compressed using an exact compact SVD calculated from its LoRA A/B factors without materializing the full dense weight updates.

The process:

1. Treats every attention, MLP and token-refiner projection independently.
2. Separates fused QKV adapters into their original Q, K and V projections.
3. Calculates the exact singular spectrum of each effective B @ A update.
4. Selects a dynamic rank independently for every projection.
5. Preserves one additional safety rank beyond the Frobenius-energy threshold.
6. Rebuilds balanced factors as A = sqrt(S) × Vh and B = U × sqrt(S).
7. Re-fuses Q, K and V using concatenated A factors and a block-diagonal B matrix.
8. Bakes the source alpha / rank scale into lora_B.
9. Removes the no-longer-required alpha tensors.
10. Saves all output tensors as BF16.

The three original v1.0/v0.1 dynamic-rank files target the same projection-level average rank:

• Rank sum: 6778
• Projection count: 312
• Average projection rank: 21.724359
• Filename rank: floor(21.724359) = 21

Q, K and V are counted separately. The stored rank of a fused QKV adapter is the sum of its three independently selected ranks.

Resize results

The Ref2V source was reduced from approximately 1,865.57 MiB to 311.79 MiB, an 83.29% reduction. The v1.1 768p source (1,865.57 MiB) was reduced to 891.65 MiB (−52.21%) at rank 64 and to 375.55 MiB (−79.87%) at rank 28.

A separate Frobenius threshold was solved for each LoRA because their singular-value distributions differ. Reusing one fixed threshold would not produce the same average rank.

Ref2V numerical validation

Direct comparison between the original and resized effective LoRA updates produced:

These are numerical weight-reconstruction measurements, not a perceptual video-quality score.

FL2V 4-step v1.1 768p numerical validation

Direct comparison between the v1.1 source and each resized variant produced:

These are numerical weight-reconstruction measurements, not a perceptual video-quality score.

Structural validation

Every resized LightX2V file contains:

Validation confirmed:

• all tensors contain finite BF16 values;
• every LoRA A tensor has a matching LoRA B tensor;
• ComfyUI recognizes all 208 adapters and all 416 tensors;
• no incompatible or unresolved tensors remain;
• QKV rank columns remain block-diagonal and ordered as Q, K and V;
• the original effective LoRA scale is preserved;
• the exact average projection rank is 21.724359.

---

Workflow Example

The following output was generated with the recommended pruned-model workflow using the v4 step-600 EMA LoRA.

<video src="https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI/resolve/main/assets/MiniMax_H3_audio.mp4" controls loop width="100%"> </video>

Current workflow

Download the current ComfyUI workflow

• Workflow: fl_minimax_h3_turbo_lora_example_workflow.json
• LoRA: minimax_h3_turbo_v4_step600_ema_pruned_comfyui.safetensors
• Steps: 6–8
• Sampler: Euler
• Scheduler: Beta
• LoRA strength: 1.0
• Video sigma shift: 12
• Audio sigma shift: 4–6

The previous workflow remains available here:

Download the previous workflow

View the previous workflow output

---

Included Files

HyperFlow 8-step LoRAs

Recommended manual HyperFlow sigmas:

1.0, 0.931506, 0.839236, 0.703462, 0.5, 0.296538, 0.160764, 0.068494, 0.0

Original HyperFlow project:

• Video-Rebirth/hyperflow — GitHub
• videorebirth/hyperflow — Hugging Face

LightX2V dynamic-rank LoRAs

Pruned/curve-form compatibility conversions

Workflows and assets

EMA and non-EMA files contain different tensor values and are retained separately.

---

Pruned/Curve-Form Compatibility Conversions

The older full-model Turbo LoRAs use keys such as:

blocks.0.attn.qkv_proj.lora_A.weight

ComfyUI expects:

diffusion_model.blocks.0.attn.qkv_proj.lora_A.weight

The compatibility conversion applies:

• blocks.* → diffusion_model.blocks.*
• token_refiner.* → diffusion_model.token_refiner.*

The original full-model LoRAs also contain AdaLN adapters whose dimensions do not match the pruned/curve-form MiniMax-H3 checkpoint. Those incompatible adapters were removed.

The removed adapters consist of 50 AdaLN projection pairs from blocks.0 through blocks.49 and one pair from final_layer.adaln_proj.linear.

These files are partial compatibility conversions and should not be assumed to behave identically to the original full-model LoRA used with its intended complete MiniMax-H3 base model.

The LightX2V dynamic-rank LoRAs do not use this pruning process.

HyperFlow is provided in separate full and pruned variants. Use the variant intended for the corresponding MiniMax-H3 base rather than mixing full-model and pruned-model files.

---

Usage

Place the LoRA files in:

ComfyUI/models/loras/

Load the selected LoRA with its compatible MiniMax-H3 model and workflow.

Recommended starting strength:

1.0

If adjustment is needed, test approximately:

0.8–1.2

When comparing files, keep the following identical:

• prompt;
• seed;
• resolution;
• frame count;
• base model;
• conditioning inputs;
• sampler;
• scheduler;
• LoRA strength;
• sigma shifts;
• workflow.

HyperFlow manual sigmas

For the four HyperFlow 8-step ComfyUI files, use:

1.0, 0.931506, 0.839236, 0.703462, 0.5, 0.296538, 0.160764, 0.068494, 0.0

The list contains nine sigma points for eight denoising steps.

---

Sampling Guidance

Three steps

For the experimental TaoMate FL2VA 3-step EMA LoRA, the manual recommended sigma sequence is:

1.0, 0.961165, 0.853333, 0.0

Four steps

Maximum speed and useful for rapid previews. Use the matching four-step LightX2V FL2V or Ref2V workflow.

For the experimental TaoMate LoRA when running four steps, the manual recommended sigma sequence is:

1.0, 0.970874, 0.907249, 0.640000, 0.0

Six steps

A good speed-to-quality compromise for the v4 step-600 EMA pruned conversion.

Eight steps

Recommended when prioritizing quality, motion consistency and synchronized-audio stability.

For the existing pruned-model workflow, start with:

• Steps: 8
• Sampler: Euler
• Scheduler: Beta
• LoRA strength: 1.0
• Video sigma shift: 12
• Audio sigma shift: 4–6

For the HyperFlow 8-step ComfyUI conversions, use the recommended manual sigma schedule:

1.0, 0.931506, 0.839236, 0.703462, 0.5, 0.296538, 0.160764, 0.068494, 0.0

HyperFlow upstream project:

• Video-Rebirth/hyperflow
• videorebirth/hyperflow

Follow the source project recommendations when using the LightX2V LoRAs:

lightx2v/Minimax-h3-Turbo

---

Audio and Scheduler Notes

MiniMax-H3 generates video and synchronized audio as separate streams with different scheduling requirements.

Incorrect audio scheduling may cause:

• severe distortion;
• noise-like output;
• unstable or blown-out audio;
• completely broken audio.

If audio is unstable, check:

• sampler and scheduler configuration;
• audio and video sigma shifts;
• manual sigma schedule when using a LoRA that requires one;
• ComfyUI version;
• MiniMax-H3 sampling implementation;
• custom sampler nodes;
• whether the workflow matches the selected FL2V, Ref2V, FL2VA or Ref2VA LoRA.

Broken audio should not automatically be attributed to the LoRA.

---

Optional Acceleration

These LoRAs may be used with compatible ComfyUI acceleration methods such as:

• Comfy Kitchen attention;
• SageAttention;
• Sol Attention;
• Spectrum.

These methods are optional and are not included in this repository. Some attention backends replace one another rather than stacking, and patch order depends on the implementation.

For few-step Turbo and HyperFlow workflows, establish a clean baseline before adding approximate attention or trajectory-forecasting methods.

---

Attribution

HyperFlow

The HyperFlow source weights and original implementation were released by Video Rebirth:

• Video-Rebirth/hyperflow — GitHub
• videorebirth/hyperflow — Hugging Face

Credit for the original HyperFlow weights, training/distillation work, scheduler design and upstream implementation belongs to Video Rebirth and its contributors.

The HyperFlow files in this repository are ComfyUI-oriented conversions and resized/pruned variants of that upstream work. They should not be confused with the original upstream HyperFlow weight file or its official loader.

The four HyperFlow files included here are:

• minimax_h3_hyperflow_8step_v1.0_comfyui_bf16.safetensors
• minimax_h3_hyperflow_8step_v1.0_comfyui_bf16_resized_avg_rank_20_bf16.safetensors
• minimax_h3_hyperflow_8step_v1.0_comfyui_pruned_bf16.safetensors
• minimax_h3_hyperflow_8step_v1.0_comfyui_pruned_bf16_resized_avg_rank_20_bf16.safetensors

Recommended manual ComfyUI sigma schedule:

1.0, 0.931506, 0.839236, 0.703462, 0.5, 0.296538, 0.160764, 0.068494, 0.0

LightX2V LoRAs

The dynamically resized FL2V and Ref2V LoRAs are derived from the official ComfyUI-compatible files released by:

lightx2v/Minimax-h3-Turbo

Credit for the source weights, training and release belongs to the LightX2V project and its contributors.

This repository provides only the dynamic-rank BF16 resizing, packaging and validation described above.

Original Turbo LoRAs

The older Turbo LoRA weights, training work and distillation method were released by:

larryvrh/MiniMax-H3-Turbo-Lora

Original ComfyUI Turbo custom node:

ComfyUI-MiniMax-H3-Turbo

This repository does not claim authorship of:

• the original HyperFlow source weights;
• the original HyperFlow training or distillation work;
• the original LoRA weights;
• LoRA training or distillation;
• the few-step sampling methods;
• MiniMax-H3 video/audio scheduling;
• the MiniMax-H3 base model;
• the pruned/curve-form checkpoint;
• the LightX2V source weights;
• the original project documentation or generation code.

The modifications provided here are limited to:

• ComfyUI-oriented HyperFlow conversions;
• pruned/curve-form HyperFlow compatibility variants;
• resized average-rank HyperFlow variants;
• pruned/curve-form ComfyUI compatibility conversions;
• namespace adjustments;
• removal of incompatible AdaLN adapters where required;
• exact dynamic-rank SVD resizing of the LightX2V ComfyUI LoRAs;
• BF16 safetensors packaging and validation.

Please retain attribution and comply with the licenses and terms of the original HyperFlow, LightX2V, Turbo LoRA and MiniMax-H3 projects when using or redistributing these files.