MiniMax H3 SageAttention Dual Mode

AI-KSK ComfyUI workflow: MiniMax H3 SageAttention Dual Mode.

This workflow is the SageAttention comparison route. It runs a unified dual-mode graph: one FL2VA branch and one Ref2VA branch, both on the 20-step quality-baseline schedule, with a KJNodes PathchSageAttentionKJ node patching the attention implementation. One relevant-looking node is not connected to the output and should not be described as active.

Main features:
- SageAttention patch via KJNodes PathchSageAttentionKJ
- Unified FL2VA and Ref2VA dual-mode graph
- 20-step simple schedule as the quality baseline
- MiniMax H3 FL2VA and Ref2VA INT8 models with Qwen3-VL encoder
- Video and audio VAE decode to saved MP4 output

Suggested workflow:
Use this route to see what SageAttention changes at the same 20-step budget. Run the same prompt and reference image as the unpatched baseline so the difference is attributable to the attention implementation, not the input.

RunningHub workflow (try online, no install needed):
https://www.runninghub.ai/zh-cn/post/2098333733807288323?inviteCode=rh-v1111

AI-KSK RunningHub workflow hub:
https://www.runninghub.ai/user-center/1892566146468511746?inviteCode=rh-v1111

Bilibili tutorial:
https://www.bilibili.com/video/BV16DYL6FEsh/

RunningHub registration and AI tools:
https://www.runninghub.ai/?inviteCode=rh-v1111
https://www.runninghub.ai/ai-generator/2046514150500524034?inviteCode=rh-v1111
https://www.runninghub.ai/vip-rights/3?inviteCode=rh-v1111
https://rhtv.runninghub.ai/projects?inviteCode=rh-v1111

Stable global access:
https://lelians.com/#/register?code=mFIJ5tZn

TokenRhythm AI model credits:
https://tokenrhythm.studio/i/rf_tr_kM6wekDquDDZQD4PUNmgiuHA

Youyun Cloud GPU:
https://passport.compshare.cn/register?referral_code=A7fTQUYY9lCDFJwA6MMvoq
https://www.notion.so/2b72459a3ac6808f8282f2c687ed8f52

RunPod Cloud GPU:
https://runpod.io/?ref=sxk5n5m7

Support AI-KSK:
https://ko-fi.com/aiksk