MiniMax H3 Image · 6-step 16-megapixel reconstruction.
A ComfyUI workflow that rebuilds a blurry or low-resolution photo into a sharp ~16MP result in 6 sampling steps: the H3 Image edit chain is conditioned on the source image, then re-sampled and decoded at high resolution.
What it is for
- Restoring blurry photos / screenshots / compressed stills
- One-click upscale-to-rebuild rather than plain interpolation
- Batch frames pulled from video
How to use
1. Load the JSON into ComfyUI (H3 Image nodes installed).
2. Feed the source image; keep the default 6-step schedule unless you need a different detail level.
3. Run; the SaveImage node writes the 16MP result.
Known limits
- It reconstructs, it does not invent detail that is not implied by the input; heavy compression artifacts may be re-interpreted.
- Needs the H3 Image model set (four weights listed in the README).
- Not a video model: process frames individually.
Episode: https://youtu.be/gnz22jMIK0Q
Chinese original: https://www.bilibili.com/video/BV1bSs76FEVm/
Workflow experience page: https://www.runninghub.ai/zh-cn/post/2121694628231987202/?inviteCode=rh-v1111
Description
MiniMax H3 Image · 6-step 16-megapixel reconstruction.
A ComfyUI workflow that rebuilds a blurry or low-resolution photo into a sharp ~16MP result in 6 sampling steps: the H3 Image edit chain is conditioned on the source image, then re-sampled and decoded at high resolution.
What it is for
- Restoring blurry photos / screenshots / compressed stills
- One-click upscale-to-rebuild rather than plain interpolation
- Batch frames pulled from video
How to use
1. Load the JSON into ComfyUI (H3 Image nodes installed).
2. Feed the source image; keep the default 6-step schedule unless you need a different detail level.
3. Run; the SaveImage node writes the 16MP result.
Known limits
- It reconstructs, it does not invent detail that is not implied by the input; heavy compression artifacts may be re-interpreted.
- Needs the H3 Image model set (four weights listed in the README).
- Not a video model: process frames individually.
Episode: https://youtu.be/gnz22jMIK0Q
Chinese original: https://www.bilibili.com/video/BV1bSs76FEVm/
Workflow experience page: https://www.runninghub.ai/zh-cn/post/2121694628231987202/?inviteCode=rh-v1111
