[V1 ] Image TO Video (For Quality Generation)
Added New Awesome Node called Load Image + Crop - This Node can crop images at real time Looks like official Node
( If Having Audio Issue Disable Lora )
[V9 Update] For Quality Generation
For Those users Who Prefer Quality With Speed (30% Slower AS compare to v7 & v8
( If Having Audio Issue Disable Lora )
⚡ [V8 Update] Lightning Speed
Now 30% Faster as Compare To V7
[V7 Update]
Welcome to the V7 Pipeline Update! This version is heavily optimized for raw speed and precision. By combining advanced 4-step architectures, this workflow is now significantly faster than any other setup currently available, delivering a massive 15% performance boost over the previous V6 workflow.
🚀 What's New in This Version?
🎭 Full Face Consistency Achieved: Integrated an advanced 4-step pipeline specifically tuned to lock in facial features perfectly across video generations.
🎬 Video-to-Video (V2V) Node Integration: Fully added and optimized the V2V node for seamless video transformations.
⚡ Extreme Prompt Enhancement (4-Step Magic): Paired a brand new 4-Step LoRA with a 4-Step Base Model. The result? Absolute maximum prompt adherence and jaw-dropping generation speeds.
⏱️ Slow-Motion Bug Fixed: Eliminated the slow-mo stuttering issues from previous versions for buttery smooth video output.
📝 Embedded Resources: The required LoRA links and exclusive rendering tips are now directly added inside the workflow nodes for easy access.
⚠️ Important Usage Tip (Please Read!)
To keep this workflow running at blistering speeds, memory management is key.
Audio Bug Warning: If you generate continuously without clearing memory, an audio bug may occur.
The Fix: Highly recommended to Clear Cache every 4 to 5 generations. This keeps your VRAM fresh and prevents audio/system glitches.
📊 Performance Benchmark
Speed: 15% Faster overall generation time compared to V6.
Efficiency: Drastically reduced VRAM load thanks to the optimized 4-step LoRA + Model combo.
🪙 Bitcoin (BTC) Wallet Address:
bc1qq6pdvukmdysef00lsgwl32xzrxut8rn8jlnusp
💀 MiniMax H3 Extreme-Fast Workflow v6.0 💀
🔥New Comfy Kitchen & Redesign UI
Running with 5 Steps
🚀 MiniMax H3 Ultra-Fast Workflow v5.0 🚀
Major Update – Speed, Clarity & Precision Redefined🔥 Major Enhancements & Key Features
⚡ True 4-Steps Lightning Performance
Extreme optimization for rapid generation without sacrificing output fidelity.
🎨 Dual LoRA System Integration
Enhanced Sharpness: Crystal-clear rendering of distant background objects and fine details.
Zero Overhead: Delivers superior visual clarity without adding extra execution steps.
🎵 Crystal-Clear Audio Processing
Integrated dedicated node processing via ComfyUI-PlagueKind-Nodes for pristine, interference-free audio output.
⚡ 50% Faster VAE Decoding (INT8 Quantizations
Upgraded VAE decode logic with INT8 precision conversion.
Cuts decoding time by 50% compared to traditional BF16 VAE Decode operations.
🧠 SageAttention Triton Optimization
SageAttention parameters are now offloaded to Triton backend execution.
Dramatically improves context processing, dynamic prompt adherence, and fine-grained scene understanding.
🔍 Next-Gen RTX Upscaler Engine
Ultra-Low Resolution Enhancer: Transforms low-density 0.2 MP renders to look as crisp and sharp as 0.5 MP native outputs.
Ensures ultra-fast image scaling with zero blur or artifacting.
📚 Integrated Node-by-Node Video Guides
Built-in YouTube tutorial links embedded directly into every key node for seamless setup and troubleshooting.
⚠️ System Note & Performance Recommendation
Resolution Scaling Advice: Because this workflow pushes maximum image quality and sharpness directly at lower base resolutions, setting extraordinarily high native output dimensions may lead to longer render times. For optimal balance between speed and quality, keep base generations lower and let the built-in RTX Upscaler handle the high-detail pass.
💖 Support the Project & Future Updates
Creating, testing, and optimizing high-performance workflows takes extensive time and compute power. To help me acquire upgraded hardware and keep bringing major updates to the community, consider supporting the work with a crypto donation!
🪙 Bitcoin (BTC) Wallet Address:
bc1qq6pdvukmdysef00lsgwl32xzrxut8rn8jlnusp
Thank you for supporting open-source AI innovation! Enjoy the extreme speed of MiniMax H3 v5.0!
About this version
in version 4.0
"The final creation: This could be my last workflow. Depends On Feedback"
⚡ WHAT'S NEW IN THIS MAJOR UPDATE ⚡
✅ Custom Fast-Moded Spectrum
└► Powered by MiniMax H3 for ultra-fast sampling! 🚀
✅ New 8-Step LoRA Integrated
└► minimax_h3_turbo_v4_step600_ema_pruned_comfyui.safetensors 🎯
✅High-Speed Performance
└► Faster & optimized CLIP loading time ⏱️
✅ Brand New UI
└► Redesigned sleek layout for a smoother experience ✨
📌 Pro Tip:
Use v3 for maximum facial consistency and detail.
Use v4 if you need the perfect balance of Medium Speed & Consistency.
in version 3.0
🔮 Key Features:
🔥 ✅ 10 Steps Generation Is Now Possible With Same Full Consistency
Just Added Turbo Lora ).
✅ Patch Sol-Attn New Alternative of Sageattention Speed Up The Workflow but in 2nd Generation
✅ MiniMax H3 Mem Eff Sage Attention Patch : To Reduce Peak Vram Usage
in version 2.0
🔮 Key Features:
🔥 Realtime Preview is Now Possible
Just Enable Node Model Preview Override ).☄️Fast Generation 0% Quality Loss : Added Agressive Spectrum & Sage Attention
Added 8 Reference Images (Note 2 or more referance can take extra time)
All Models & Nodes & Tutorial Added (Direct links)
Easy To Use Workflow
Face Consistency is on Ultra Level even on 0.2 Megapixels
Added Low Vram GPU Mathes
Description
in version 3.0
🔮 Key Features:
🔥 10 Steps Generation In Now Possible With Same Full Consistency
Just Added Turbo Lora ).Patch Sol-Attn New Alternative of Sageattention Speed Up The Workflow but in 2nd Generation
MiniMax H3 Mem Eff Sage Attention Patch : To Reduce Peak Vram Usage
FAQ
Comments (32)
No Image 2 Video ? Only Ref 2 video ?
RTX 5080 16gb vram + 32 gb ram
115 seconds for a 6 sec video, 0.4 megapixels. Cool
127 for an 8 second video.
But I don't really understand why my characters mumbles jibberish in the first seconds and then speaks as intended. I used my previous prompt, anything you've encountered?
Edit: Also, are you gonna add upscaling?
are you using the turbo lora with <20 steps? it was added to the v3 of this workflow, and it hasn't been great for audio.
@wondier Ah, I'll try it out if that's it. Even with the turbo lora, I'm getting some videos with ok speech, like with the Joker/batman interactions.
Tried the workflow with 8GB VRAM and 32GB RAM. Used 0.4 megapixels.
Duration= 8s -> (OOM)
Duration= 5s -> (OOM)
Duration= 3s -> (Success)
Is this the expected result?
can tell me what GPU you are using, do you updated your comfy ui ? its really good working even on 3050 with upto 8sec 0.4 mega pixel.
bro, I have a 4060ti graphics card with 8GB. It records 1-megapixel videos of 5-6 seconds without any problems.
@Voxe1 what is your generation time for 5-6 sec video with your vram?
Are you on Windows or Linux? Windows is far less prone to cuda oom issues in my experience.
@fakolonya Right now, I mainly make 3‑second videos; if the prompt works well, I make them 5 seconds long. 3‑second video (generation time: 5 minutes) 5‑second video (generation time: 12 minutes) I made a 6‑second video once, and it took about 20 minutes.
@FloatsYourStoat Windows 11
大佬,这个工作流里的那些attention加速可以叠加吗
Thank you for this, runs well on 4070 12GB, 64GB system ram. Had to disable the preview and the sageattention nodes because wheels for linux aren't available (yet).
ComfyUI-Easy-Install comes with a 1 click install bat for sage attention and it work fine in windows. https://github.com/Tavris1/ComfyUI-Easy-Install
not sure if that helps with your problem?
SageAttention-Multi - Installs both SageAttention v2.2.0 and v3 (v3 effective only on NVIDIA 50-series GPUs)
@Wurstibert This! I got tired of doing multiple Comfy installs so often due to updates so I started using ComfyUI Easy Install too. 1 click Comfy install then another for sage. I got so lazy I made a batch file to remove folders under /model and create symbolic links to /model folders I have on another SSD.
https://huggingface.co/Kijai/PrecompiledWheels/blob/main/sageattention-2.2.0-cp312-cp312-linux_x86_64.whl
is the wheel you will need for linux.
Sage Attention is easy to compile in Linux, no need for pre-build wheels
@bubblegum1 Unless they updated it nope, won't build without changing a certain line of code in some file I forget, other wise it fails the checks and does not build. And they removed the pre built wheel for linux for sage attention 2.2.0
is anyone else having backgrounds randomly swap into other backgrounds?
bro its reference to video not - image to video,you can add node for it, its so easy
@RedditUser9811 @hatt2 Technically you can prompt it so reference image dont change. But its kinda unstable. Really wish you can add a reference like audio on I2V node
@RedditUser9811 thats a good point. will try. thanks!
At 0.5 megapixels, my RTX5060ti took 325 seconds to generate a 10-second video clip. That's acceptable and very good !!
I have an RTX 4090 GPU and 96GB of RAM, so why am I still getting "insufficient VRAM" errors? I can only run at 0.2x megapixels for a 5-second video; furthermore, whenever I load "minimax_h3_turbo_4step," it reports insufficient VRAM regardless of the video resolution settings.
Do you get this error with this workflow only or with others too?
I don't know man, what OS are you running on and what are your command line parameters?
Check your dedicated GPU settings. Update your driver. Maybe you are running with integrated graphics by mistake.
@jasssingh1121379 +1
It's your settings. I'm running on 4070ti super with 32gb ram and 4090 with 64gb ram. No issues, runs fast af. Easily generating 25 second videos at .7 mp with ~10 minute generation time. It scales even faster if I used less references and shorter video. Are you running this on windows or linux?
Hi I am having audio issues in this, initial words are getting missed, How do I fix this?
For me the generations are working fast and good enough, if you are able to tell which optimization node is causing this, I will just bypass/disable it, I can afford to take the 20% speed hit for the audio fix
its manybe due to low steps sometimes sound crash or not working we need a good lighting lora
Sorry to bother you, but I can't seem to find where to set the video duration
in location where you enter prompt look at bottom-right