This two-pass MiniMax H3 Ref2VA workflow focuses on multi-reference video redraw. It is intended for scenes where character identity, clothing, props, environment, or style must be reinforced with several images while the generated video is enlarged and refined in a second pass.
The graph combines MiniMax H3 FL2VA and Ref2VA through the H25 hybrid loader, uses Qwen3-VL 32B MiniMax encoding, and includes the MiniMax H3 reference LoRA at strength 0.75. Multiple image inputs feed the reference-conditioning route. Pass one uses an eight-step beta schedule, after which the video is enlarged and reconstructed with a four-step simple schedule, Euler sampling, and the detail-oriented sigma-shift settings.
The workflow is packaged as a ready-to-run ComfyUI graph for users who want to compare generative enlargement against conventional video scaling. Keep the original source, prompt, and duration unchanged when comparing versions so that visible differences come from the workflow structure rather than from changed creative inputs.
Main features:
- Two-stage Ref2VA multi-reference redraw
- H25 hybrid FL2VA and Ref2VA model loader
- Multiple image reference slots
- MiniMax H3 reference LoRA at strength 0.75
- Eight-step beta first pass
- RTX enlargement between generation stages
- Four-step Euler refinement pass
- 16:9 reference-to-video conditioning route
Suggested workflow:
Give every reference image a single clear role and avoid loading near-duplicate references with conflicting details. Establish identity in the first two slots, then use later slots for clothing, props, or environment. Keep the second-pass prompt conservative and focused on clarity. If references compete with one another, remove the least important image before changing sampler parameters.
RunningHub Workflow
Try the workflow online right now - no installation required.
Workflow: https://www.runninghub.ai/post/2089197212697116673?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 & Asia-Pacific)
Watch the workflow demonstration and creative breakdown here:
Bilibili Video: https://www.bilibili.com/video/BV1kZbQ6xEMg/
Support Me on Ko-fi
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/2089197212697116673?inviteCode=rh-v1111
如果觉得效果理想，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 48 GB 大显存性能。
B站视频：https://www.bilibili.com/video/BV1kZbQ6xEMg/
我会在夸克网盘持续更新模型资源：
https://pan.quark.cn/s/07bdc81784ce
Description
This two-pass MiniMax H3 Ref2VA workflow focuses on multi-reference video redraw. It is intended for scenes where character identity, clothing, props, environment, or style must be reinforced with several images while the generated video is enlarged and refined in a second pass.
The graph combines MiniMax H3 FL2VA and Ref2VA through the H25 hybrid loader, uses Qwen3-VL 32B MiniMax encoding, and includes the MiniMax H3 reference LoRA at strength 0.75. Multiple image inputs feed the reference-conditioning route. Pass one uses an eight-step beta schedule, after which the video is enlarged and reconstructed with a four-step simple schedule, Euler sampling, and the detail-oriented sigma-shift settings.
The workflow is packaged as a ready-to-run ComfyUI graph for users who want to compare generative enlargement against conventional video scaling. Keep the original source, prompt, and duration unchanged when comparing versions so that visible differences come from the workflow structure rather than from changed creative inputs.
Main features:
- Two-stage Ref2VA multi-reference redraw
- H25 hybrid FL2VA and Ref2VA model loader
- Multiple image reference slots
- MiniMax H3 reference LoRA at strength 0.75
- Eight-step beta first pass
- RTX enlargement between generation stages
- Four-step Euler refinement pass
- 16:9 reference-to-video conditioning route
Suggested workflow:
Give every reference image a single clear role and avoid loading near-duplicate references with conflicting details. Establish identity in the first two slots, then use later slots for clothing, props, or environment. Keep the second-pass prompt conservative and focused on clarity. If references compete with one another, remove the least important image before changing sampler parameters.
RunningHub Workflow
Try the workflow online right now - no installation required.
Workflow: https://www.runninghub.ai/post/2089197212697116673?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 & Asia-Pacific)
Watch the workflow demonstration and creative breakdown here:
Bilibili Video: https://www.bilibili.com/video/BV1kZbQ6xEMg/
Support Me on Ko-fi
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/2089197212697116673?inviteCode=rh-v1111
如果觉得效果理想，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 48 GB 大显存性能。
B站视频：https://www.bilibili.com/video/BV1kZbQ6xEMg/
我会在夸克网盘持续更新模型资源：
https://pan.quark.cn/s/07bdc81784ce