<div data-youtube-video="https://youtu.be/SUJkkT6XLHI"><iframe src="https://www.youtube.com/embed/SUJkkT6XLHI" width="853" height="480" allowfullscreen></iframe></div> 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2086411670364184577?inviteCode=rh-v1111
<h2>MiniMax H3 Reference-To-Video — 48 GB Edition</h2>
A MiniMax H3 video graph from episode 291, part of the six-times-acceleration series that compares model tier, LoRA, attention and cache settings.
<h2>Saved configuration</h2>
27 nodes / 34 links. LoadImage + LoadVideo + LoadAudio. UNET <code>minimax_h3_ref2va_int8_convrot.safetensors</code>; CLIP <code>qwen3vl_32b_minimax_h3_nvfp4_awq</code>; int8 video/fp32 audio VAEs; turbo-v4 LoRA at 1; euler + beta 6 steps denoise 1.
<h2>How to use</h2>
The 48 GB reference edition: int8-convrot ref2va weights and nvfp4-awq CLIP. Same six-step turbo recipe as the 24 GB file; the difference is the quantization tier.
<h2>Limits and not-verified items</h2>
No inference was run for this listing: no speed, VRAM, quality, audio-sync or acceleration-factor claim is made. The "6x" in the episode title is the author's description and was not independently measured. Values are quoted from the saved nodes. H3 weights, LoRAs, encoder and VAEs are not bundled; check their licenses separately. The graph notes credit the official Comfy-Org/MiniMax-H3 model files; preserved as-is, no originality claim. The cover image is a promotional still, not a frame produced by this graph.
Tutorial: <a href="https://www.youtube.com/watch?v=SUJkkT6XLHI" rel="ugc">YouTube</a> · <a href="https://www.bilibili.com/video/BV17nuU6xE3g" rel="ugc">Bilibili</a>. Workflows supplied through AI-KSK; verify workflow, node, model, LoRA and cover rights separately. No blanket commercial or redistribution license is asserted.

MiniMax H3 Reference-to-Video 6x Speedup - 48 GB Edition

<div data-youtube-video="https://youtu.be/SUJkkT6XLHI"></div> 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2086411670364184577?inviteCode=rh-v1111
MiniMax H3 Reference-To-Video — 48 GB Edition
A MiniMax H3 video graph from episode 291, part of the six-times-acceleration series that compares model tier, LoRA, attention and cache settings.
Saved configuration
27 nodes / 34 links. LoadImage + LoadVideo + LoadAudio. UNET <code>minimax_h3_ref2va_int8_convrot.safetensors</code>; CLIP <code>qwen3vl_32b_minimax_h3_nvfp4_awq</code>; int8 video/fp32 audio VAEs; turbo-v4 LoRA at 1; euler + beta 6 steps denoise 1.
How to use
The 48 GB reference edition: int8-convrot ref2va weights and nvfp4-awq CLIP. Same six-step turbo recipe as the 24 GB file; the difference is the quantization tier.
Limits and not-verified items
No inference was run for this listing: no speed, VRAM, quality, audio-sync or acceleration-factor claim is made. The "6x" in the episode title is the author's description and was not independently measured. Values are quoted from the saved nodes. H3 weights, LoRAs, encoder and VAEs are not bundled; check their licenses separately. The graph notes credit the official Comfy-Org/MiniMax-H3 model files; preserved as-is, no originality claim. The cover image is a promotional still, not a frame produced by this graph.
Tutorial: <a href="https://www.youtube.com/watch?v=SUJkkT6XLHI" rel="ugc">YouTube</a> · <a href="https://www.bilibili.com/video/BV17nuU6xE3g" rel="ugc">Bilibili</a>. Workflows supplied through AI-KSK; verify workflow, node, model, LoRA and cover rights separately. No blanket commercial or redistribution license is asserted.

07_MiniMax_H3_____6___48GB_____1_.json

v1.0

https://youtu.be/SUJkkT6XLHI 
 
This workflow is the 48GB high-configuration reference-to-video acceleration version for MiniMax H3. It is designed for stronger GPU setups where the reference-driven route has more memory headroom than the 24GB general version. 
 
The active chain includes a MiniMax H3 Ref2VA INT8 ConvRot model, Qwen3-VL MiniMax NVFP4/AWQ text encoding, MiniMax H3 video and audio VAEs, MiniMaxH3ReferenceToVideo control, turbo LoRA acceleration, and SaveVideo output. 
 
This package is meant for practical ComfyUI and RunningHub users who want a ready-made MiniMax H3 graph instead of rebuilding loader, encoder, VAE, sampler, and video export connections by hand. The workflow is especially useful for speed comparison, hardware planning, and repeatable prompt tests. Keep the same source image and prompt when comparing versions so the difference you see comes from the workflow route, not from changing creative variables. 
 
Main features: 
 
- MiniMax H3 reference-to-video workflow for 48GB-class hardware 
- Reference-to-video control node is part of the output path 
- INT8 ConvRot Ref2VA model route with Qwen3-VL MiniMax encoding 
- MiniMax H3 video and audio VAE components 
- Euler sampling with beta 6-step schedule 
- 16:9 horizontal video output 
- Middle-tier reference workflow between 24GB compatibility and 80GB full configuration 
 
Suggested workflow: 
 
Use this after the 24GB reference workflow has confirmed the reference direction. Keep the same reference material, prompt, duration, and seed strategy, then compare whether the 48GB route holds identity, lip motion, and camera movement more consistently. 
 
For better comparisons, make one small test first, review identity stability and temporal motion, then raise prompt complexity only after the base route behaves correctly. When using speaking-character prompts, keep the face visible, avoid too many fast cuts, and separate action, expression, and camera instructions clearly. 
 
RunningHub Workflow 
 
Try the workflow online right now - no installation required. 
Workflow: https://www.runninghub.ai/post/2086424051652087809?inviteCode=rh-v1111 
 
If the results meet your expectations, you can later deploy it locally for customization. 
 
Fan Benefits: Register to get 1000 points + daily login 100 points - enjoy 4090 performance and 48 GB super power! 
 
Bilibili Updates (Mainland China &amp; Asia-Pacific) 
 
If you're in the Asia-Pacific region, you can watch the video below to see the workflow demonstration and creative breakdown. 
Bilibili Video: https://www.bilibili.com/video/BV17nuU6xE3g/ 
 
Support Me on Ko-fi 
 
If you find my content helpful and want to support future creations, you can buy me a coffee. 
Every bit of support helps me keep creating. 
Ko-fi: https://ko-fi.com/aiksk 
 
Business Contact 
 
For collaboration or inquiries, please contact aiksk95 on WeChat. 
 
打开下方链接即可在线体验，无需安装。 
工作流：https://www.runninghub.ai/post/2086424051652087809?inviteCode=rh-v1111 
 
如果你觉得效果理想，也可以在本地进行自定义部署。 
 
粉丝福利：注册即可领取 1000 积分，每日登录再领 100 积分，体验 4090 和 48GB 大显存性能。 
 
B站视频（中国大陆及亚太地区） 
 
如果你在中国大陆或亚太地区，可以通过下面的视频查看工作流的实测效果与创作思路。 
B站视频：https://www.bilibili.com/video/BV17nuU6xE3g/ 
 
我会在夸克网盘持续更新模型资源： 
https://pan.quark.cn/s/07bdc81784ce

Files with SHA256: bf072875a26dc4753b84d215bc77b488f59d8f9aaa070bec93f8f5b0c2e186e0