MiniMax-H3 INT8 FLF, R2V, Bridge & TTS Workflows Collection

Optimized ComfyUI workflows for MiniMax-H3, featuring INT8 inference, Sage-Attention, SLA-Attention, Spectrum, Lightx2v LoRA, Turbo-LoRA, native Extend support, Face Refine, Text-to-Speech, Latent Upscaler integration for faster high-resolution generation, Bridge video generation, and Group Bypass control.

Workflows

Workflow	File	Description
First-Last-Frame (FLF)	`MiniMax_int8_I2V-javanoYYMM.x.json`	FLF video generation with native Extend support and Latent Upscaler integration.
Reference-to-Video (R2V)	`MiniMax_int8_R2V-javanoYYMM.x.json`	Reference-guided video generation with a dedicated reference LoRA, native Extend support, and Latent Upscaler integration.
Bridge	`MiniMax_int8_Bridge-javanoYYMM.x.json`	Generates a new transition video between Video A and Video B, then connects the two videos into a continuous sequence.
Face Refine	`MiniMax_int8_FR-javano2608.1.json`	Face refinement workflow for correcting and enhancing faces in generated video.
Text-to-Speech (TTS)	`MiniMax_int8_TTS-javanoYYMM.x.json`	High-speed Text-to-Speech using MiniMax-H3 with audio-only output.

The Latent Upscaler and SLA-Attention are integrated directly into the I2V and R2V workflows.

Base model:

Comfy-Org/MiniMax-H3
https://huggingface.co/Comfy-Org/MiniMax-H3

Features

INT8

INT8 MiniMax-H3 inference for:

Reduced VRAM usage
Faster inference
Improved memory efficiency

Sage-Attention

Optimized attention implementation for faster MiniMax-H3 inference and improved generation throughput.

ComfyUI-SolAttn_triton
https://github.com/kijai/ComfyUI-SolAttn_triton

SLA-Attention

ComfyUI-H3-SLA-Attention
https://github.com/PlagueKind/ComfyUI-PlagueKind-Nodes/tree/main/ComfyUI-H3-SLA-Attention

SLA-Attention is integrated into the I2V and R2V workflows to provide an additional optimized attention implementation for MiniMax-H3 inference.

The implementation is intended to improve inference efficiency and reduce the computational cost of attention operations.

Spectrum

ComfyUI-Spectrum-MiniMax-H3
https://github.com/xmarre/ComfyUI-Spectrum-MiniMax-H3

Optimized sampling and MiniMax-H3 inference acceleration.

Lightx2v LoRA

lightx2v/Minimax-h3-Turbo
https://huggingface.co/lightx2v/Minimax-h3-Turbo

Acceleration LoRA for reduced sampling steps and faster MiniMax-H3 video generation.

Turbo-LoRA

MiniMax-H3-Turbo-Lora-ComfyUI
https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI

Experimental low-step acceleration. Output quality may vary depending on LoRA strength and sampling settings.

Native Extend

MiniMax-H3 supports video extension natively, allowing existing video content to be extended directly within the workflow.

This can be used to continue generated videos while maintaining temporal and visual continuity.

Latent Upscaler

ComfyUI MiniMax-H3 Latent Upscaler
https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler

The MiniMax-H3 Latent Upscaler has been integrated directly into the I2V and R2V workflows.

It uses a two-stage generation process:

Generate the initial video at a lower resolution.
Upscale the latent representation to a higher resolution before decoding.

By generating the initial video at a lower resolution, the computational cost of the main generation stage is reduced. The latent representation is then upscaled to the target resolution, allowing high-resolution output with a shorter overall generation time.

The Latent Upscaler is currently integrated into:

I2V / FLF workflow
R2V workflow

Recommended for:

High-resolution video generation
Faster generation
Lower-resolution initial sampling
More efficient I2V workflows
More efficient R2V workflows
Reducing the computational cost of direct high-resolution generation

The actual speed improvement depends on the input/output resolution, number of frames, sampling configuration, GPU, and other workflow settings.

Bridge

The Bridge workflow generates a new video that connects two existing video clips.

It takes Video A and Video B as inputs, generates a new transition sequence between them, and then combines the results into a continuous video:

Video A → Generated Bridge → Video B

This is useful for connecting two separate cuts while creating a generated transition that matches the visual content of both clips.

Recommended for:

Connecting two separate video cuts
Creating seamless transitions
Bridging different scenes
Extending visual continuity between clips
Creating transitions between existing generated videos
Combining independently generated video segments

Group Bypass Control

ComfyUI-Switchboard
https://github.com/Oratorian/ComfyUI-Switchboard

Used in the workflows to control Group Bypass states.

The Switchboard Group Controller allows workflow groups to be enabled or bypassed, making it possible to switch optional processing paths and workflow configurations without manually modifying the graph.

R2V Reference LoRA

The R2V workflow requires the following reference LoRA:

minimax_h3_ref_lora_rank_256_bf16.safetensors

https://huggingface.co/Kijai/MiniMax-H3-experimental/tree/main/loras

This LoRA provides the reference-image conditioning required by the R2V workflow.

Face Refine

ComfyUI-H3-FaceRefine
https://github.com/Carasibana/ComfyUI-H3-FaceRefine

Provides automatic face refinement for generated video, similar to a Face Detailer workflow.

Features:

Automatic face detection
Face refinement
Facial detail enhancement
Improved facial consistency
Video face correction

Text-to-Speech

MiniMax-H3 can also be used as a dedicated Text-to-Speech engine.

The TTS workflow generates audio only, bypassing video generation entirely. This significantly reduces processing requirements and enables extremely fast speech synthesis.

Workflow Details

First-Last-Frame (FLF)

`MiniMax_int8_I2V-javanoYYMM.x.json`

Generates controlled transitions between a first and last frame.

Features:

First + Last Frame conditioning
Native Extend support
Latent Upscaler
SLA-Attention
Group Bypass control
INT8
Sage-Attention
Spectrum
Lightx2v LoRA
Optional Turbo-LoRA

Recommended for:

Cinematic transitions
Character animation
Camera movement
Transformations
Keyframe-based video generation
High-resolution video generation with reduced generation time

Latent Upscaler Integration

The I2V workflow can generate the initial video at a lower resolution and then upscale the latent representation before decoding.

This allows the main video generation stage to operate at a lower resolution, reducing the computational cost compared with directly generating the final high-resolution video.

SLA-Attention Integration

SLA-Attention is integrated into the I2V workflow as an optimized attention option for MiniMax-H3 inference.

Reference-to-Video (R2V)

`MiniMax_int8_R2V-javanoYYMM.x.json`

Generates video from a reference image with dedicated reference LoRA conditioning and native Extend support.

Features:

Reference image conditioning
Reference LoRA
Character/appearance consistency
Native Extend support
Latent Upscaler
SLA-Attention
Group Bypass control
INT8
Sage-Attention
Spectrum
Lightx2v LoRA
Optional Turbo-LoRA

Recommended for:

Character videos
Consistent subjects
Appearance preservation
Image-driven animation
Reference-guided video generation
High-resolution reference-based video generation

Latent Upscaler Integration

The R2V workflow can generate the initial video at a lower resolution and then upscale the latent representation to the target resolution before decoding.

This reduces the computational cost of the initial video generation stage and can significantly shorten generation time for high-resolution R2V generation.

SLA-Attention Integration

SLA-Attention is integrated into the R2V workflow as an optimized attention option for MiniMax-H3 inference.

Required R2V LoRA

minimax_h3_ref_lora_rank_256_bf16.safetensors

Download from:

https://huggingface.co/Kijai/MiniMax-H3-experimental/tree/main/loras

Place the LoRA in the appropriate ComfyUI LoRA directory and load it in the R2V workflow.

Bridge

`MiniMax_int8_Bridge-javanoYYMM.x.json`

Generates a new transition video between two existing video clips and combines them into a continuous sequence.

Processing Flow

Video A → Generated Bridge → Video B

The workflow uses the ending portion of Video A and the beginning portion of Video B to generate a new intermediate video sequence.

The generated Bridge video is then inserted between the two source videos, producing a connected final sequence.

Features

Video A input
Video B input
Generated transition video
Connection between separate video cuts
Continuous video output
INT8
Turbo-LoRA

Recommended for

Connecting two independent video cuts
Scene transitions
Shot-to-shot transitions
Creating intermediate motion
Bridging different generated video segments
Improving continuity between separate clips

The Bridge workflow is particularly useful when two independently generated clips do not connect naturally. Instead of simply concatenating the clips, MiniMax-H3 generates an intermediate sequence designed to connect the visual states of the two clips.

Face Refine

`MiniMax_int8_FR-javano2608.1.json`

A dedicated face refinement workflow for MiniMax-H3 generated video.

Uses:

ComfyUI-H3-FaceRefine
https://github.com/Carasibana/ComfyUI-H3-FaceRefine

The workflow automatically detects and refines faces in the generated video, providing a Face Detailer-like correction process.

Recommended for:

Facial detail enhancement
Correcting facial artifacts
Improving face consistency
Refining generated characters

Text-to-Speech (TTS)

`MiniMax_int8_TTS-javanoYYMM.x.json`

A dedicated Text-to-Speech workflow using MiniMax-H3.

The workflow outputs audio only, without generating video frames. By bypassing the video generation pipeline, it enables extremely fast speech synthesis with significantly lower processing requirements.

Features:

Text-to-Speech using MiniMax-H3
Audio-only output
INT8 inference
Fast speech generation
ComfyUI workflow integration

Recommended for:

Voice generation
Dialogue creation
Voice-over production
Rapid audio prototyping
Generating speech for video production

Latent Upscaling

MiniMax-H3 Latent Upscaler

Repository:
https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler

Model:
https://huggingface.co/LBH-123-AI/Minimax_h3_latent_Upscaler

The MiniMax-H3 Latent Upscaler is integrated into the I2V and R2V workflows.

It is designed to reduce generation time when producing high-resolution video by separating the video generation and resolution-upscaling stages.

Processing Flow

Low-resolution generation → Latent Upscaling → High-resolution decoding

The initial video is generated at a lower resolution, which reduces the computational cost of the main sampling process.

The resulting latent representation is then upscaled to the desired resolution before the final VAE decoding stage.

Advantages

Faster initial video generation
Lower computational requirements during the main generation stage
More efficient high-resolution generation
Reduced sampling cost compared with direct high-resolution generation
Faster workflow iteration

The Latent Upscaler is currently integrated into:

MiniMax_int8_I2V-javanoYYMM.x.json
MiniMax_int8_R2V-javanoYYMM.x.json

SLA-Attention

ComfyUI-H3-SLA-Attention

Repository:
https://github.com/PlagueKind/ComfyUI-PlagueKind-Nodes/tree/main/ComfyUI-H3-SLA-Attention

SLA-Attention is integrated into the I2V and R2V workflows.

It provides an optimized attention implementation for MiniMax-H3 and can be used to improve inference efficiency.

The current workflows use SLA-Attention specifically in the I2V and R2V workflows.

Group Bypass Control

ComfyUI-Switchboard

Repository:
https://github.com/Oratorian/ComfyUI-Switchboard

The workflows use ComfyUI-Switchboard for Group Bypass control.

The Group Controller is used to enable or bypass specific workflow groups, allowing optional processing paths and workflow configurations to be switched without manually changing node connections.

Required Custom Nodes

ComfyUI-SolAttn_triton
https://github.com/kijai/ComfyUI-SolAttn_triton
ComfyUI-Spectrum-MiniMax-H3
https://github.com/xmarre/ComfyUI-Spectrum-MiniMax-H3
ComfyUI-H3-SLA-Attention
https://github.com/PlagueKind/ComfyUI-PlagueKind-Nodes/tree/main/ComfyUI-H3-SLA-Attention
ComfyUI-H3-FaceRefine
https://github.com/Carasibana/ComfyUI-H3-FaceRefine
ComfyUI MiniMax-H3 Latent Upscaler
https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler
ComfyUI-Switchboard
https://github.com/Oratorian/ComfyUI-Switchboard

Install through ComfyUI Manager → Install Missing Custom Nodes or python -m pip install according to the installation instructions of each custom node.

Required Models

Main Model

Comfy-Org/MiniMax-H3
https://huggingface.co/Comfy-Org/MiniMax-H3

R2V Reference LoRA

minimax_h3_ref_lora_rank_256_bf16.safetensors
https://huggingface.co/Kijai/MiniMax-H3-experimental/tree/main/loras

Required for the R2V workflow.

Lightx2v LoRA

lightx2v/Minimax-h3-Turbo
https://huggingface.co/lightx2v/Minimax-h3-Turbo

Used for accelerated video generation with reduced sampling steps.

Optional Turbo-LoRA

MiniMax-H3-Turbo-Lora-ComfyUI
https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI

Used for experimental low-step acceleration.

Latent Upscaler

Minimax_h3_latent_Upscaler
https://huggingface.co/LBH-123-AI/Minimax_h3_latent_Upscaler

Required when using the Latent Upscaler integrated into the I2V or R2V workflows.

Installation

Update ComfyUI and ComfyUI-Manager.
Install the required custom nodes listed above.

You can install the custom nodes using either:
- ComfyUI Manager → Install Missing Custom Nodes
- python -m pip install, following the installation instructions provided by each custom node.
Download Comfy-Org/MiniMax-H3.
For R2V, download:
- minimax_h3_ref_lora_rank_256_bf16.safetensors
Download lightx2v/Minimax-h3-Turbo if using Lightx2v acceleration.
Download MiniMax-H3-Turbo-Lora-ComfyUI if using Turbo-LoRA.
Download Minimax_h3_latent_Upscaler if using Latent Upscaler functionality in the I2V or R2V workflows.
Load the desired workflow:
- MiniMax_int8_I2V-javanoYYMM.x.json — FLF / I2V
- MiniMax_int8_R2V-javanoYYMM.x.json — R2V
- MiniMax_int8_Bridge-javanoYYMM.x.json — Bridge
- MiniMax_int8_FR-javano2608.1.json — Face Refine
- MiniMax_int8_TTS-javanoYYMM.x.json — TTS

Recommended Configuration

High-Speed Configuration

INT8 + Sage-Attention + Turbo-LoRA (4 steps)

INT8 + Sage-Attention + Spectrum + Turbo-LoRA (8 steps)

For high-resolution I2V and R2V generation:

Low-resolution generation → Latent Upscaler → High-resolution output

For Bridge generation:

Video A → MiniMax-H3 Bridge generation → Video B

These configurations reduce the computational cost of video generation while providing fast generation and high-resolution output when Latent Upscaling is used.

The optimal configuration may vary depending on:

GPU
Input resolution
Target resolution
Number of frames
Sampling steps
LoRA strength
Desired output quality

Notes

FLF = First-Last-Frame.
R2V = Reference-to-Video.
Bridge generates an intermediate video between Video A and Video B and connects the two source videos.
FR = Face Refine.
TTS = Text-to-Speech.
Extend is natively supported by MiniMax-H3.
The Latent Upscaler is integrated into the I2V and R2V workflows.
SLA-Attention is integrated into the I2V and R2V workflows.
The R2V workflow requires minimax_h3_ref_lora_rank_256_bf16.safetensors.
The TTS workflow outputs audio only and does not generate video.
Latent Upscaling uses low-resolution generation followed by latent upscaling to reduce the computational cost of high-resolution generation.
ComfyUI-Switchboard is used for Group Bypass control in the workflows.
Turbo-LoRA is experimental.
Sage-Attention and Spectrum require compatible environments.
SLA-Attention requires a compatible environment.
Combining multiple acceleration methods may require tuning of sampling steps and LoRA weights.
If audio is not generated correctly, try lowering the shift_audio value of the ModelSamplingMiniMaxH3 node inside the INT8 Model Loader subgraph from 4.00 to 3.00.

Motion Context Compatibility Note

The ComfyUI-H3-Motion-Context custom node is no longer required by the latest workflows.

Starting with:

I2V: 2608.9
R2V: 2608.16

Motion Context has been replaced with the native MiniMaxH3AddGuide node. Therefore, ComfyUI-H3-Motion-Context is no longer required for the latest workflow versions.

Additional Resources

MiniMax-H3
https://huggingface.co/Comfy-Org/MiniMax-H3

R2V Reference LoRA
https://huggingface.co/Kijai/MiniMax-H3-experimental/tree/main/loras

Sol-Attn
https://github.com/kijai/ComfyUI-SolAttn_triton

SLA-Attention
https://github.com/PlagueKind/ComfyUI-PlagueKind-Nodes/tree/main/ComfyUI-H3-SLA-Attention

Spectrum MiniMax-H3
https://github.com/xmarre/ComfyUI-Spectrum-MiniMax-H3

H3 Face Refine
https://github.com/Carasibana/ComfyUI-H3-FaceRefine

Lightx2v LoRA
https://huggingface.co/lightx2v/Minimax-h3-Turbo

MiniMax-H3-Turbo-Lora-ComfyUI
https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI

MiniMax-H3 Latent Upscaler
https://github.com/LBH-123-AI/Comfyui_Minimax_h3_latent_Upscaler

MiniMax-H3 Latent Upscaler Model
https://huggingface.co/LBH-123-AI/Minimax_h3_latent_Upscaler

ComfyUI-Switchboard
https://github.com/Oratorian/ComfyUI-Switchboard

License

Please follow the respective licenses of MiniMax-H3, ComfyUI, the listed custom nodes, LoRAs, the Latent Upscaler, SLA-Attention, and other third-party components used by these workflows.

Downloads last month: -