MiniMax H3 Continuum
GitHub:
https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum
[ V3.5 ]
V3.5 introduces two major additions:
- Continuum-aware Second Pass / Hi-Res Fix
Refine externally processed H3 latents while preserving Continuum physical groups, prompts, seeds, ordering, and first-pass audio. An integrated one-node 2x Hi-Res Fix path is also included as an experimental feature.
- Low-memory Assemble + Seam V3.5
Adds Auto, RAM, and Disk-backed video-buffer modes. Disk-backed assembly significantly reduces system RAM/private-memory usage for long or high-resolution outputs while preserving Exact Duration, Seam, Terminal Merge, and audio behavior.
All V3.4 nodes remain available for saved-workflow compatibility. Existing V3.4 workflows continue to work unchanged.
The V3.5 release passed 430 automated tests and representative GPU acceptance tests.
Note: The integrated Hi-Res Fix remains experimental. Long 2x workflows can require substantial GPU VRAM.
#5 3 chunks x 15 seconds
#6 6 chunks x 15 seconds
W576xH576
[ v3.4 ]
Long-form MiniMax H3 video and audio generation for ComfyUI with chunked generation, persistent references, restartable runs, and user-controlled audio.
### What's new in v3.4
- Driving Audio: preserves the supplied audio as the final audio while guiding generation across chunks.
- Video Reference: provides persistent visual reference for identity, motion, framing, and scene appearance.
- Restartable chunks: reuse completed chunks with Run Storage and regenerate only the required part.
- Improved Core compatibility: unknown upstream or custom nodes are not rejected merely because they are not recognized by Continuum.
- Simpler stable interface: obsolete compatibility controls and experimental Timeline inputs are hidden from the V3.4 public workflow.
- Spectrum interoperability: Spectrum remains optional and can use the official H3 Continuum Interop API.
### Direction change from v3.3
V3.4 focuses on predictable reference workflows rather than experimental Timeline Video and timeline-audio generation.
Driving Audio preserves the original user-supplied audio. Video Reference provides persistent visual guidance without requiring exact frame-by-frame copying. Existing V3.3 workflows remain available through legacy compatibility paths.
### Updating
For an existing Git installation:
git pull --ff-only origin mainV3.4 input connection patterns
V3.4 separates the visual reference input from the driving-audio input. Choose the connection pattern that matches your source material.
1. Audio only
Connect Load Audio to driving_audio. Use this when an existing song, dialogue track, or sound effect should remain the final audio. A Video Reference is not required.
2. Video with its own audio
Connect Load Video (Upload) IMAGE to Video Reference. If the uploaded video contains the audio you want to preserve, connect its AUDIO output to driving_audio as well.
3. Video and audio from separate sources
Connect Load Video (Upload) IMAGE to Video Reference, then connect a separate Load Audio node to driving_audio. Use this when the visual reference video and the final audio source are different files.
Both inputs are optional. Connect Video Reference when visual guidance is needed, and connect driving_audio when the supplied audio should be preserved in the final output.
Video Reference frame rate
Use a 24 fps source for Video Reference. Load Video (Upload) may accept files recorded at 25 fps or another frame rate, but acceptance alone does not guarantee correct temporal alignment with H3. For a non-24 fps source, set force_rate to 24 in Load Video (Upload), or convert the file to 24 fps before loading it. If the source is already 24 fps, leave force_rate at its default and do not resample it.
Current validation status
[ v3.3 ]
V3.3 adds Timeline Video conditioning for long-form MiniMax H3 generation. A reference video can now be processed in chunk-local time slices, allowing motion and scene continuity to be carried across multiple 5-second chunks while keeping the reference resolution independent from the output resolution. The Efficient 0.4 MP mode helps reduce memory usage and processing time.
Video assembly has also been improved. Auto seam handling analyzes chunk boundaries and applies guarded corrections for transient flicker, micro-flash, exposure, and color differences. This helps produce more natural transitions between generated chunks without changing the original sampling process.
Existing V3.2.4 workflows remain available as Legacy nodes for compatibility.
[ v3.24 ]
Generate longer native MiniMax H3 video and audio sequences in ComfyUI.
H3 Continuum is a ComfyUI custom node that generates a longer sequence as connected chunks and assembles them into one continuous video.
```text
3 × 5-second chunks → 15-second video
6 × 5-second chunks → 30-second video
The previous video and audio latent context is passed into each continuation chunk. This is not a simple video concatenation workflow.
Main purpose: longer MiniMax H3 generation, not faster generation.
Easy Installation
H3 Continuum can be installed directly from ComfyUI Manager.
Open ComfyUI Manager
Search for H3 Continuum or Continuum
Select Install
Restart ComfyUI
Load one of the included sample workflows
Manual installation and the latest documentation are available on GitHub:
GitHub:
https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum
What It Does
H3 Continuum divides a longer generation into manageable chunks.
MiniMax H3 Model
↓
H3 Continuum Sampler
↓
ComfyUI Core Video / Audio VAE Decode
↓
H3 Continuum Assemble
↓
Final videoEach continuation chunk receives latent context from the preceding chunk. Overlapping context is removed during assembly, and the final frame and audio counts are aligned to the requested duration.
Main Features
Connected long-form MiniMax H3 generation
Native video and audio latent continuation
Fixed, List, and Timeline prompt formats
Automatic prompt-format detection
T2VA, I2VA, FL2VA, Last Frame and Reference workflows
Up to three Reference Images
Reference Audio conditioning
First Frame and Last Frame conditioning
Configurable continuity context
Run Storage and automatic resume
Partial regeneration from a selected chunk
Optional Spectrum interoperability
Standard and Turbo sample workflows
ComfyUI Core VAE Decode compatibility
Included Sample Workflows
Two example workflows are provided.
Standard Workflow
Recommended when output quality and temporal consistency are the priority.
Standard MiniMax H3 sampling
Spectrum can be enabled
Suitable for quality-focused generation
Reference Image and Reference Audio supported
RTX upscaling can be enabled when required
Turbo Workflow
Recommended for faster tests and iteration.
LightX2V MiniMax H3 Turbo LoRA
8-step example configuration
Spectrum is bypassed by default
Faster than the standard workflow in tested configurations
Some loss of facial detail or additional artifacts may occur
Turbo LoRA models:
https://huggingface.co/lightx2v/Minimax-h3-Turbo/tree/main
MiniMax H3 models and documentation:
https://huggingface.co/MiniMaxAI/MiniMax-H3
Models and LoRAs are not included with this custom node.
Reference + Continuation
Reference Images remain available across all generated chunks.
A typical setup is:
Picture 1 → face and identity
Picture 2 → full-body appearance and clothing
Picture 3 → environment or an additional visual reference
Audio 1 → vocal, music or audio-performance referenceRef2VA is the reference-specialized checkpoint and is generally the first choice for stronger reference fidelity.
FL2VA with Reference conditioning is also allowed. H3 Continuum does not automatically replace or switch the connected model.
Spectrum Integration
Spectrum is optional. H3 Continuum also works without it.
With a compatible Spectrum release, H3 Continuum sends a continuation signal only when generating later chunks.
Chunk 1 → normal Spectrum sampling
Chunk 2+ → Continuum Actual Prefix 2This allows Spectrum to coordinate its spectral forecasting with the continuation context instead of treating every chunk as an unrelated generation.
Benefits include:
Automatic identification of continuation chunks
Actual Prefix applied only where required
No manual prefix switching between chunks
Reduced risk of duplicated prefix processing
Compatibility with standard ComfyUI workflow execution
Spectrum remains an approximate accelerator. Motion, anatomy, audio and detail can differ from a non-Spectrum result, so quality comparisons should use the same prompt and seed.
Spectrum:
https://github.com/xmarre/ComfyUI-Spectrum-MiniMax-H3
Run Storage and Resume
Enable Save + Auto Resume to preserve completed raw video and audio chunks.
If a generation is interrupted, H3 Continuum can reuse compatible saved chunks and continue from the first missing chunk.
It can also regenerate from a selected chunk while preserving the compatible prefix.
Chunk 1–3 completed
↓
Generation interrupted
↓
Queue the workflow again
↓
Chunks 1–3 reused
↓
Generation continues from Chunk 4Run Storage verifies the sampling contract, model route, references, resolution and saved chunk files before reuse.
Prompt Formats
Fixed
One prompt is used for every chunk.
List
Separate prompts are divided with:
---Timeline
[0-5s]
First scene description
[5-10s]
Second scene description
[10-15s]
Third scene descriptionPrompt Format = Auto detects the appropriate format automatically.
Incomplete timeline coverage produces diagnostics and safe fallback behavior rather than unnecessarily stopping every generation. Structurally unusable input is still reported as an error.
Tested Configuration
The current Windows implementation has been tested with:
GPU NVIDIA RTX 5060 Ti 16GB
ComfyUI MiniMax H3-compatible Core build
Chunk Duration 5 seconds
Typical Length 3 or 6 chunks
Continuity Balanced 22 frames
Standard Sampling RES Multistep
Spectrum Interop Actual Prefix 2The node is not limited to RTX 50-series GPUs. Actual compatibility, generation speed and usable resolution depend on the MiniMax H3 model, GPU memory, ComfyUI configuration and installed acceleration nodes.
RTX 4060 and other configurations have not been formally validated by this project.
Frequently Asked Questions
Is this only a workflow?
No. H3 Continuum is a ComfyUI custom node package. The included workflows are ready-to-use examples.
Does it generate one native 30-second sample?
No. It generates connected chunks and assembles them into one longer output while carrying video and audio latent context forward.
Does it make MiniMax H3 faster?
Speed is not the primary purpose. H3 Continuum is designed for longer generation. Spectrum and Turbo LoRAs can reduce generation time in some configurations.
Is Spectrum required?
No. It is an optional acceleration and interoperability path.
Can I use the Turbo LoRA?
Yes. A Turbo sample workflow is provided. Spectrum is bypassed by default in that workflow because combining both can change quality or introduce artifacts.
Which model should I use for Reference Images?
Ref2VA is the reference-specialized option. FL2VA with Reference conditioning is also allowed, but reference fidelity may differ.
Are the models included?
No. MiniMax H3 checkpoints, text encoders, VAEs, Turbo LoRAs and optional acceleration nodes must be installed separately.
Can an interrupted generation be resumed?
Yes. Enable Run Storage before generation. Compatible completed chunks can then be reused.
Can I regenerate only the later part?
Yes. Run Storage supports regeneration from a selected chunk while retaining a compatible earlier prefix.
Are chunk boundaries always invisible?
No generative continuation system can guarantee a completely invisible boundary. H3 Continuum preserves latent context and removes duplicated overlap, but difficult motion, lighting changes and large prompt transitions can still produce flicker or visual changes.
Does Reference Audio guarantee exact lip synchronization?
Reference Audio conditions MiniMax H3’s native joint video/audio generation. It can guide vocals, rhythm, expression and mouth movement, but it does not guarantee sample-identical audio reproduction or frame-perfect lip synchronization in every generation.
Does it support audio continuity?
Yes. Video and audio latent context are carried together. The assembler also provides an optional Audio Seam mode for boundary-local audio correction.
Is RTX 5090 required?
No. Development and runtime validation were performed on an RTX 5060 Ti 16GB. Lower-memory configurations may require reduced resolution, offloading or other ComfyUI memory optimizations.
What license is used?
H3 Continuum is released under the MIT License.
Links
Custom Nodes
The included Standard and Turbo workflows use the following custom nodes.
- H3 Continuum
https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum
- rgthree-comfy
https://github.com/rgthree/rgthree-comfy
- ComfyUI-Easy-Use
https://github.com/yolain/ComfyUI-Easy-Use
- ComfyUI-KJNodes
https://github.com/kijai/ComfyUI-KJNodes
- ComfyUI-Spectrum-MiniMax-H3
https://github.com/xmarre/ComfyUI-Spectrum-MiniMax-H3
- NVIDIA RTX Nodes for ComfyUI
https://github.com/Comfy-Org/Nvidia_RTX_Nodes_ComfyUI
Spectrum and RTX upscaling are optional generation paths, but installing all listed custom nodes allows the included workflows to load without missing-node warnings.
Models
- MiniMax H3
https://huggingface.co/MiniMaxAI/MiniMax-H3
- LightX2V MiniMax H3 Turbo LoRA
https://huggingface.co/lightx2v/Minimax-h3-Turbo/tree/main
Models and LoRAs are not included in the workflow ZIP.
Main Links
- GitHub and documentation
https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum
- Install from ComfyUI Manager
Search for H3 Continuum
Description
V3.3 introduces Timeline Video conditioning for chunk-based long-form generation.
Video Seam handling has also been improved. Auto mode analyzes chunk boundaries and applies guarded corrections to reduce visible transitions between chunks.
FAQ
Comments (29)
version 3.4 is broken. Even tried on a fresh installation of comfyui
Thank you for reporting this. You were correct: the initial V3.4 repository package was incomplete because v3/driving_nodes.py was missing. I apologize for the broken release.
This has now been fixed in commit 84c22ff.
Please update the H3 Continuum custom node with git pull, or reinstall/update it through ComfyUI Manager, and then restart ComfyUI.
If the error remains after updating, please share the startup console error so I can check it immediately.
@ukr8b3g201 I really appreciate your quick response and hard work. Special thanks again.
Thank you for reporting this. You were correct: the initial V3.4 package was incomplete, which caused an API mismatch between the V3.4 node and the older sampler implementation. This was not caused by your workflow or prompt.
I apologize for the broken release. I am correcting the package and verifying a clean installation before publishing the fixed version.
Not being able to change the prompt for a chunk without starting over completely is a deal breaker. Unless the video is extremely simple, it's almost impossible to provide a perfect prompt the first time.
Thank you for the valuable feedback. I agree that being able to adjust prompts for individual chunks without restarting the entire generation would be very useful, especially for longer or more complex videos.
I'll consider possible approaches for a future version. It may require some changes to how Continuum manages chunk state and resuming, so I can't promise an implementation yet, but it's definitely something worth exploring.
Testing with 5s chunks helps, but seeds seem very impactful still
@Xarfai I agree — this is an important limitation for longer or more complex generations.
At the moment, Run Storage can preserve completed chunks and resume compatible runs, but the current public V3.4 interface does not yet expose a reliable way to select a specific chunk, change its prompt, and regenerate from that point.
So for now, changing a prompt in the middle of a sequence may still require restarting more of the generation than is ideal.
This is something I want to improve. The goal is to make it possible to keep the good earlier chunks, revise the prompt for a later section, and regenerate from that point without throwing away the whole sequence.
the seed behavior is also important here, so I’m testing both prompt revision and seed handling before treating this as a finished feature.
为什么我生成3段5秒视频，最后出来的视频，不是连续的，有时候会从头开始
这种情况仅凭目前的信息还很难判断具体原因。
如果只是使用一个比较简单的提示词来生成多个 5 秒片段，模型有时可能会在后续片段中重新开始类似的动作，而不是自然地延续前一个片段。
建议尝试使用 Timeline（时间线）或 List（列表）形式的提示词，对不同时间段 / Chunk 分别描述动作和剧情的发展，例如明确指定每个阶段人物应该继续做什么，而不是让所有片段重复使用相同的描述。
也可以参考 MiniMax H3 官方的提示词指南来设计时间线和动作描述。
如果方便的话，也可以提供你实际使用的 Prompt 和 Workflow / 设置，这样会更容易判断是提示词的问题，还是 Continuum 的连接或生成逻辑出现了问题。
我用0-5秒 5-10秒 10-15秒来写的提示词哦，好像并没有暗写的来展示
Looove the workflow, been having great ref2va results with the XUELUO Checkpoint at 15s and 8step lora. 1 minute is very consistent with it, more minutes depend a bit on the seed.
Therefore if I could wish for sth, it would be a "feature" that some SVI workflows have, which would be being able to generate n steps in the timeline, view them and then continue on, if youre happy.
Still amazing, but just an idea.
Thanks for the detailed feedback — this is a very constructive suggestion.
I like the idea of being able to generate a few chunks, review the result, and then continue from there instead of committing to the entire timeline at once.
Continuum already has some of the underlying resume/regeneration mechanisms, so I’d like to explore how this kind of step-by-step workflow could fit into a future version. I can’t promise when or in what form it will be implemented yet, but this is definitely a direction worth investigating.
It's amazing how well the audio and video sequences are coordinated. T2V itself generates beautifully. My advice for creating a cohesive video—for example, if we have four 10-second segments for a 40-second video—is to format the prompts like this:
[0-10s]
prompt
[10-20s]
prompt
[20-30s]
prompt
[30-40s]
prompt
This produces stunning results.
Thank you for sharing this. It is especially useful to know that four explicit 10-second sections worked well for a cohesive 40-second video, including audiovisual coordination.
So far, most of my validation has used 5-second chunks, but your result suggests that longer chunk durations can also work effectively when each section is clearly defined with matching time ranges. I will include this as a useful prompting example and investigate 10-second chunks more thoroughly.
@ukr8b3g201 it works with 15s chunks as well, Ive generated i2v of up to 4 minutes (16x15s).
Scene consistency sometimes got lost a bit, but currently I assume this due to prompting since I get those issues with shorter combinations too. Guess i need to practice more :)
But as far as getting the videos generated and progressing as expected, I would say it works well.
@Xarfai 返信案：
Thank you, that is extremely useful feedback. A 4-minute I2VA generation using 16 × 15-second chunks is an impressive result and confirms that Continuum is not limited to 5-second chunks.
Your observation about scene consistency is also valuable. Since similar drift can occur with shorter chunks, prompting, source material, model, and seed may be more significant factors than chunk duration alone.
I will add 10- and 15-second chunk examples to the documentation and test them more systematically. Thank you for sharing the exact configuration and result.
@ukr8b3g201 if you want me to test anything like this or give you feedback, im always open disc [at]xarfai
@Xarfai may i ask in which workflow you achieved that? Standard, Turbo? T2V or I2V?
and which quality settings?
i've tested the turbo workflow yesterday and had some issues with my first runs (15s x 6 chunks) until i lowered the quality to 0.5 and chunks to 5.
it then generated the 5 chunks but still every second or third run the wf gave me an H3 Continuum Assemble + Seam V3.4 error.
@denolim465778 I did turbo workflow 8 step with 1 Megapixel quality on the xueluo checkpoint. was ref2va with 3 images, 39 frames overlap and keep identity, trying multiple NSFW loras, so i wont include them, but should just make it more stable
@Xarfai Thank you, I really appreciate the offer.
Your 16 × 15s / 1MP result is especially useful because it gives me a good real-world reference for longer Continuum runs. I’m currently testing 10s and 15s chunks more systematically, including FL2VA and long Ref2VA sequences.
If I find a specific case where I need an external test, especially around 15s chunks, 39-frame overlap, long-duration consistency, or regeneration/resume behavior, I’ll definitely reach out.
Also, could you let me know what GPU you are using and how much system RAM you have? That would help me compare your results with different hardware setups.
Thanks again for sharing the exact settings and for offering to help test.
@ukr8b3g201 Have a 5090 and 96Gb of RAM
@Xarfai say whaaat?
I did T2V with the Turbo WF.
8 step with 0.4 Megapixel on minimax_h3_fl2va_pruned_int8_convrot.safetensors.
All i could get was 5x15s with some OOM every second run.
This on a 5090 with 96 GB RAM.
Ok. I'm confused now. :-)
The model I use is 19GB. yours 37GB.
And you dont get any OOMs when generating a 4 minute video?
Do you own a Datacenter or something? :-)
i really don't get it. Lol :-)
Thanks a lot btw. i will try xueluo today.
I am still confused :-)
@Xarfai you make me hope :-)
@denolim465778 That is actually very useful information, especially now that we know you both have a 5090 and 96 GB of RAM.
The two tests are quite different though, so I would not compare them only by checkpoint size.
Your test was T2V with the FL2VA pruned INT8 ConvRot model at 0.4 MP, while Xarfai used Ref2VA with the Xueluo checkpoint, 3 reference images, 39-frame overlap and Keep Identity. ComfyUI can also move parts of the workload between VRAM and system RAM, so checkpoint file size alone does not directly tell us how much memory the complete run will require.
What interests me more is that you are getting OOMs mainly after repeated runs. That could indicate memory not being fully released between generations, possibly during VAE decode or final assembly, rather than the video length alone.
If it happens again and it is convenient, an error log would be very helpful. The easiest ways are:
Copy the last 20–30 lines from the ComfyUI console/terminal immediately after the error, especially the traceback and any CUDA out of memory message.
If ComfyUI shows the error directly on the failed node, copying or screenshotting that error is also fine.
If the Continuum Status / Report output contains additional information for that run, you can paste that as well.
No need to collect everything — whatever is easiest is useful.
Since you and Xarfai have essentially the same GPU and RAM, comparing the two setups may actually help isolate whether the difference comes from the model/workflow path or from memory cleanup between runs.
@ukr8b3g201 it seems the model itself is big factor.
@Xarfai and i own the same hardware.
but he has a 4 minute video and i do not :-) Lol
i ran my tests on
minimax_h3_fl2va_pruned_int8_convrot.safetensors
will try xueluo today.
hope it helps
@ukr8b3g201 yes. interesting.
OK.
i will also test the model he suggested ASAP.
let's see if i will be able to make a 4 minute video ;-)
@denolim465778 hope it works out for you, also dont do RTX upscale at that length, thats the one time I got an OOM Error, which is understandable and not an issue as upscaling can be done seperately after.
@Xarfai thank you. nope. didn't. ;-)
i assume i might have some false configuration.
i saw somewhere in the logs that something expects numpy 2.4 and i have 2.5.
upscaler was always bypassed in all runs.
thank you