MiniMax H4 ComfyUI: Free Workflows, Setup, VRAM & Fixes
Run MiniMax H4 locally in ComfyUI with T2V, I2V and reference-video workflows. Check which models to download, whether your GPU can handle H3, and how to fix common VRAM and workflow problems.
MiniMax H4 is MiniMax's general-purpose multimodal video generation model. For the model specs, system architecture, and workflow guidance, see the MiniMax H4 model guide.
MiniMax H4 ComfyUI Status
Last reviewed: August 9, 2026
Native ComfyUI support: Available
Official
T2V: Available
Official
I2V: Available
Official
Reference-to-Video: Available
Official
Local H3 Base: Available
Official
Official full Regenerate-2K open weights: Not yet available
Official
Low-VRAM workflows: Community / experimental
Experimental
MiniMax H4 ComfyUI Quick Answer
Yes. MiniMax H4 can run locally in current versions of ComfyUI using native workflows for text-to-video, image-to-video and reference-video generation. The main challenge is hardware: H3 is a large video model, so usable performance depends heavily on GPU VRAM, system RAM, model precision, resolution and CPU offloading. Users with lower-VRAM GPUs can use quantized or aggressively offloaded workflows, while users who do not want to manage large model downloads or hardware limits can run MiniMax H3 online instead.
Before downloading, review which H3 components are public and which remain hosted in the H3 open weights and license guide.
Need GPU-Specific MiniMax H4 Guidance?
VRAM alone does not determine whether a MiniMax H4 workflow will run well. GPU architecture, system RAM, operating system, model format and offloading also matter. Use the dedicated hardware checker before choosing your local model files.
How Much RAM Does MiniMax H4 Need?
VRAM is not the only limit. System RAM becomes important for CPU offloading, text encoder loading, VAE work, large model movement and low-VRAM workflows. MiniMax does not publish one universal system-RAM minimum for every H3 workflow.
Check MiniMax H4 RAM and offloading guidanceWhich MiniMax H4 ComfyUI Workflow Should You Use?
Image to Video (I2V)
Animate a starting image while controlling motion, camera movement and audio through the prompt.
Reference to Video (R2V)
Use visual or video references to guide character, identity or scene consistency.
Download MiniMax H4 ComfyUI Workflows
This site does not host local workflow JSON files for this page. Use the current official ComfyUI MiniMax H4 workflow guide or the trusted ComfyUI-compatible model repository instead of downloading fabricated files.
Which MiniMax H4 Models Should I Download?
What do you want to generate?
What matters most?
Recommended model category
T2V or I2V / FL2VA workflow family
Choose the recommended ComfyUI-compatible category before experimental variants.
Check the current official ComfyUI MiniMax H4 model repository for the latest recommended file.
Check My HardwareWhere Do MiniMax H4 Models Go in ComfyUI?
ComfyUI/
└── models/
├── diffusion_models/
├── text_encoders/
└── vae/diffusion_models
MiniMax H4 diffusion model
text_encoders
H3 text encoder
vae
MiniMax H4 video VAE
How to Install MiniMax H4 in ComfyUI
Step 1 - Update ComfyUI
Use a current ComfyUI version that supports MiniMax H4. Keep this step version-flexible because support details can change.
Step 2 - Choose a Workflow
Pick T2V, I2V or R2V based on your input type.
Step 3 - Download the Required Models
Download the diffusion model, text encoder and VAE category required by the selected workflow.
Step 4 - Place the Models in the Correct Folders
Use the diffusion_models, text_encoders and vae folders shown above.
Step 5 - Import the Workflow
Drag in a workflow JSON or use the ComfyUI Template Library when available.
Step 6 - Start With a Conservative Resolution
Low-VRAM users should not begin at the highest resolution. Confirm a smaller run first.
Step 7 - Generate and Monitor Memory Usage
Watch VRAM, system RAM and OOM behavior during model load, sampling and VAE decode.
Can MiniMax H4 Generate 2K Locally in ComfyUI?
Not with the complete official open-weight 2K pipeline yet.
MiniMax H4 Base can run locally in ComfyUI, with the open local workflow primarily centered around the H3 Base generation stage. The complete official 2K pipeline involves additional stages such as H3-Regenerate-2K, and those full open weights are not yet available as of August 9, 2026.
| Capability | Local Open-Weight Status |
|---|---|
| H3 Base T2V | Available |
| H3 Base I2V | Available |
| H3 Base R2V | Available |
| ~768p-class Base workflow | Available |
| Full official Regenerate-2K stage | Not yet open-weight |
MiniMax H4 ComfyUI Performance Benchmarks
Tested by MiniMaxH4 Lab
Our benchmark dataset is being built. We only publish results after reproducing the workflow.
| GPU | VRAM | RAM | Model | Resolution | Video Length | Steps | Generation Time | Evidence |
|---|---|---|---|---|---|---|---|---|
| Pending | - | - | - | - | - | - | - | No fabricated data |
Community-Reported Results
Community-reported results are not MiniMaxH4.org benchmarks and may vary by workflow, software version and system configuration.
| GPU | VRAM | RAM | Model | Resolution | Video Length | Steps | Generation Time | Evidence |
|---|---|---|---|---|---|---|---|---|
| RTX 3060 | 12GB | High system RAM recommended | Quantized/offloaded community workflow | Lower starting resolution | Short clips | Varies | Slower than 24GB+ GPUs | Community reported; reproduce before citing |
Future MiniMaxH4 Lab tests should prioritize RTX 3060, RTX 4070 Ti Super, RTX 4090 and RTX 5090.
How to Make MiniMax H4 Faster in ComfyUI
Use SageAttention
SageAttention may improve attention performance, but results vary by hardware and workflow. Do not promise a fixed 2x speedup.
Reduce Resolution First
Resolution usually has the largest practical impact on memory and generation time.
Reduce Video Length
Shorter clips reduce compute and memory pressure.
Use Lower-Precision or Quantized Models
Lower precision can reduce VRAM usage, with possible quality or stability tradeoffs.
Use CPU Offloading on Low-VRAM GPUs
Offloading can make a workflow fit, but it may slow generation.
Close Other GPU Applications
Free VRAM before loading H3 models.
Experimental Optimization Methods
W4A8, INT8, GGUF, Turbo / acceleration LoRA and cache-based optimization are experimental / community-developed - not part of the standard official MiniMax H4 workflow.
MiniMax H4 ComfyUI Known Issues
Out of Memory
Community ReportedCause: VRAM is not enough, the resolution is too high, or model precision is too demanding for the GPU.
Fix: Lower resolution, use a quantized model, and increase offloading.
Text Encoder OOM
Community ReportedCause: The text encoder can consume substantial memory before generation begins.
Fix: Use CPU offload or a lighter encoder path when the workflow supports it.
VAE Decode OOM
Community ReportedCause: Decoding video frames can exceed memory after sampling appears to finish.
Fix: Lower resolution, reduce video length, and increase system RAM/offloading.
Missing Nodes
OfficialCause: The workflow was opened with an older ComfyUI build or missing dependencies.
Fix: Update ComfyUI and dependencies, then reload the workflow.
Dark Frames or Periodic Stutter
Community ReportedCause: Some users have reported periodic dark-frame or stutter behavior in specific MiniMax H4 ComfyUI configurations. Treat this as a community-reported issue rather than a universal H3 defect.
Fix: Test a conservative resolution, update ComfyUI, and compare against a standard native workflow before changing multiple variables.
Quantized Model Crashes
ExperimentalCause: Experimental models can be affected by GPU, CUDA, PyTorch and ComfyUI versions.
Fix: Re-test with the standard workflow or a different quantized variant before blaming the base model.
MiniMax H4 ComfyUI Troubleshooting
What problem are you seeing?
Out of memory
- Most likely cause
- The workflow is exceeding available VRAM during model load, sampling, text encoding, or VAE decode.
- First fix
- Reduce resolution and video length before changing several settings at once.
- Second fix
- Use a lower-precision or quantized workflow and enable CPU offloading.
- If it still fails
- Run the online generator when hardware tuning costs more time than the video is worth.
If the local setup keeps failing, compare the time cost against browser-based generation.
MiniMax H4 ComfyUI vs Online: Which Should You Use?
| ComfyUI Local | MiniMax H4 Online | |
|---|---|---|
| Setup | Manual | None |
| GPU required | Yes | No |
| Large model download | Yes | No |
| Workflow control | Maximum | Lower |
| Hardware troubleshooting | Required | None |
| Local privacy | Better | Cloud processing |
| Best for | Power users | Fast generation |
Use ComfyUI if you want maximum workflow control and already own suitable hardware. Use the online generator if you primarily want to create videos without downloading large models, managing VRAM or troubleshooting local dependencies.
No local GPU setup required when you choose the online path.
MiniMax H4 ComfyUI Prompt Tips
A useful prompt should identify the parts H3 needs to coordinate:
Example
A woman in a red raincoat walks through a neon-lit Tokyo alley at night, handheld tracking shot, wet pavement reflecting signs, realistic motion, distant traffic and soft rainfall ambience.
MiniMax H4 ComfyUI FAQ
Does MiniMax H4 work in ComfyUI?
Yes. MiniMax H4 works in current ComfyUI versions through native workflows for text-to-video, image-to-video and reference-to-video. The practical constraint is not only software support; local generation also depends on GPU VRAM, system RAM, model precision, resolution and offloading settings.
Which ComfyUI version supports MiniMax H4?
Use a current ComfyUI build that includes native MiniMax H4 workflow support. Avoid hard-coding a future-sensitive version number into your setup notes; update ComfyUI first, then load the native MiniMax H4 templates from the Template Library or the current official documentation.
How much VRAM does MiniMax H4 need?
MiniMax does not publish one universal minimum VRAM requirement for MiniMax H4. GPU architecture, system RAM, operating system, model format, resolution, video length and offloading all matter. Use the dedicated MiniMax H4 Local Hardware Checker for configuration-specific guidance.
Can RTX 3060 12GB run MiniMax H4?
A 10–12GB MiniMax H4 setup should be treated as experimental. Current results depend heavily on the exact GPU architecture, optimized model files, system RAM, storage, operating system and offloading strategy. Validate the exact configuration in the dedicated hardware checker before downloading model files.
Can MiniMax H4 run on 8GB VRAM?
There is not enough reproducible evidence to recommend 8GB VRAM as a practical MiniMax H4 target today. The current detailed low-memory evidence leaves 8GB unverified rather than demonstrating an end-to-end H3 workflow on a physical 8GB GPU.
Is 16GB VRAM enough for MiniMax H4?
A specific optimized MiniMax H4 ComfyUI workflow has been reproducibly measured on a 16GB RTX 5070 Ti. That proves one Blackwell, Linux and NVMe configuration can work; it does not make 16GB an official minimum or prove that every 16GB GPU behaves identically.
Can RTX 4090 run MiniMax H4?
Yes. An RTX 4090 with 24GB VRAM is one of the more practical local GPU configurations for MiniMax H4 ComfyUI workflows. It gives more room for higher resolutions and reduces the need for aggressive offloading, but workflow settings can still cause OOM errors.
Can RTX 5090 run MiniMax H4?
Yes. A 32GB RTX 5090-class GPU should provide a better local MiniMax H4 experience than 16GB or 24GB cards. That does not mean every workflow is immune to out-of-memory errors, and generation speed still depends on software, precision, resolution and video length.
How much RAM does MiniMax H4 need?
System RAM matters because CPU offloading, text encoder loading, VAE work, and large model movement can spill beyond VRAM. 32GB is a practical lower target for some optimized setups, 64GB is better for low-VRAM workflows, and 96GB or more helps with heavy experimentation.
Where do MiniMax H4 model files go in ComfyUI?
MiniMax H4 diffusion models go in ComfyUI/models/diffusion_models, the H3 text encoder goes in ComfyUI/models/text_encoders, and the video VAE goes in ComfyUI/models/vae. Check the current official ComfyUI MiniMax H4 model repository for the latest exact filenames.
Which MiniMax H4 workflow should I download?
Download the workflow that matches the input you want to use. Choose T2V for prompt-only generation, I2V when you start from an image, and R2V when references should guide identity, scene, motion or consistency. Do not download every workflow unless you actually need all modes.
Does MiniMax H4 generate audio in ComfyUI?
Yes. MiniMax H4 workflows can generate video with native stereo audio when the required audio components and workflow outputs are configured correctly. If you see silent output, check the audio VAE, export settings and player before assuming the model cannot produce audio.
Can MiniMax H4 generate 2K locally?
Not with the complete official open-weight 2K pipeline yet. H3 Base can run locally in ComfyUI, but the complete official 2K path includes additional stages such as H3-Regenerate-2K. The full Regenerate-2K open weights are not available as of August 9, 2026.
Why is MiniMax H4 so slow in ComfyUI?
MiniMax H4 is slow because video generation is compute heavy and memory movement is expensive. Low-VRAM workflows often rely on CPU offloading, which can keep the job running but add significant latency. Reducing resolution and video length usually helps more predictably than chasing experimental optimizations first.
How do I fix MiniMax H4 out-of-memory errors?
Start by reducing resolution, then reduce video length, then switch to lower precision or quantized models if needed. Enable CPU offloading on low-VRAM GPUs and close other GPU applications. If OOM appears during VAE decode, reduce output size and make sure system RAM is sufficient.
Speed up MiniMax H4 with Spectrum
Spectrum Apply MiniMax H4 can replace selected native H3 transformer evaluations with feature forecasts in supported ComfyUI workflows.
Spectrum Apply MiniMax H4 →About the author
MiniMaxH4.org Editorial publishes hands-on AI video generation workflows, MiniMax H4 prompt engineering guides, and practical testing of multimodal video tools.
Run MiniMax H4 Without Local Setup
The fastest way to decide how to run MiniMax H4 in ComfyUI is to know when not to run it locally. If model downloads, VRAM limits or dependency issues are blocking you, generate online.