<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/2086444236236484609?inviteCode=rh-v1111
<h2>MiniMax H3 Larry + Attention + Cache Full Acceleration — MiniMax H3 Larry + Attention + Cache Full Acceleration</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>
71 nodes / 86 links. Three parallel branches: UNET <code>minimax_h3_fl2va_int8_convrot</code>, CLIP <code>qwen3vl_32b_minimax_h3_int8_convrot</code>, turbo-v4 LoRA at 1, euler + beta 6 steps denoise 1; three LoadImage inputs and three outputs.
<h2>How to use</h2>
The 'full blood' acceleration file: three branches with attention and cache patches layered on the same int8 model and turbo LoRA. Like the SageAttention file, it is a comparison rig; pick one branch's patch stack if you want a single production graph.
<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 Larry + Attention + Cache Full Acceleration

<div data-youtube-video="https://youtu.be/SUJkkT6XLHI"></div> 
RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2086444236236484609?inviteCode=rh-v1111
MiniMax H3 Larry + Attention + Cache Full Acceleration — MiniMax H3 Larry + Attention + Cache Full Acceleration
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
71 nodes / 86 links. Three parallel branches: UNET <code>minimax_h3_fl2va_int8_convrot</code>, CLIP <code>qwen3vl_32b_minimax_h3_int8_convrot</code>, turbo-v4 LoRA at 1, euler + beta 6 steps denoise 1; three LoadImage inputs and three outputs.
How to use
The 'full blood' acceleration file: three branches with attention and cache patches layered on the same int8 model and turbo LoRA. Like the SageAttention file, it is a comparison rig; pick one branch's patch stack if you want a single production graph.
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.

09_MiniMax_H3_Larry_attention_cache______.json

v1.0

https://youtu.be/SUJkkT6XLHI 
 
This workflow is the full-speed MiniMax H3 acceleration version built around Larry Turbo, attention optimization, and cache-assisted execution. It is the more aggressive performance-oriented option in this batch. 
 
The connected graph keeps MiniMax H3 video generation as the core and adds the acceleration structure needed for the full-speed route. It is intended for users who already understand the normal MiniMax H3 workflow and now want to push iteration speed further. 
 
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: 
 
- Full-speed MiniMax H3 acceleration workflow 
- Larry Turbo, attention optimization, and cache-oriented execution design 
- Built for faster repeated image-to-video testing 
- Keeps the MiniMax H3 model, encoder, VAE, sampler, and video output structure together 
- Useful for comparing maximum-speed output against safer baseline routes 
- Good match for hardware testing and practical production timing checks 
- Cover and workflow are matched to the full-speed cache version 
 
Suggested workflow: 
 
Use this version after you have already validated the prompt on a normal or lower-risk acceleration path. Keep a known good source image and prompt, then test how much quality you are willing to trade for speed. For production work, save a baseline output first so you can compare temporal stability and subject identity objectively. 
 
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/2086705735132983297?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/2086705735132983297?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: a8d596dc13cb908cc846a119b2359b9ba470e6651925329c9242380a1082da73