MiniMax RefMod Studio v3.0
i added the refmod tab
so you dont need to make a custom node for refmod it does it by (comfy api)
have fun
the refmod is so sensetive to the refrences so i made this tool
some tips for refrences:
1. always use full light refrences
2. use loud and clear audios (use the tool in app to fix audio)
3. the res of the images should be the same ( otherwise it will use the first image as aspect ratio and ) ( the tool will do it automaticly for you )
4. use fullReference mode ALWAYS!
5. the res should not go blow 512 * 512
6. you are doing a character refmod use closeups also
---
One-click dataset & reference prep for images and audio
Turn a messy folder of pictures and voice clips into a clean, perfectly named, ready-to-use reference set. No coding, no setup. Just pick a folder and press Build Dataset.
Made by MHJ Studio (its my studio name not my civitai name)
---
What it does?
Images
- Resize to MiniMax-standard sizes (1024×1024, 768×1024, 1024×768) or 512 / 768 / 1536 / 720p / original / custom
- Fit (pad), Fill (crop) or Stretch
- Transparent, black, white, custom-color or blurred-image backgrounds
- Export as PNG, JPG or WEBP with a quality slider
- Auto-fix photo rotation, auto contrast, sharpen, grayscale
- Skip tiny images, skip duplicates, optionally save a mirrored copy of every image
Audio
- Copy as-is, or convert to WAV / MP3 / FLAC / OGG
- Set sample rate, downmix to mono, even out the volume
- Volume control up to 400% (with a limiter so boosted audio doesn't clip harshly)
- Denoise with Light / Medium / Strong
- FFmpeg is built in, so there is nothing extra to install
Naming & organizing
- Custom base name, separator, start number and digit padding
- Pair each image with its audio file under the same number
- Include sub-folders, auto-create a {name}_dataset folder
- Optional caption .txt file next to every image
- manifest.csv showing exactly what became what
Quality
- Live preview of your settings on a real image from your folder
- Dry run mode: see what would happen without writing anything
- Progress bar with ETA, activity log and a Cancel button
- Save / load presets
- Drag a folder onto the window
- 5 accent colors, animated aurora background, adjustable window opacity
---
How to use
1. Download and run the .exe (Windows)
2. Choose your Input folder
3. Pick an output folder, or tick the auto-folder option
4. Set your size, naming and audio options
5. Press Build Dataset ✅
---
Coming next: VIDEO REFERENCE UPDATE
Video reference support is in the works. Stay tuned!
---
⚠️ Notes
- Windows only for now.
- Your original files are never modified. Everything is written to the output folder.
- Some antivirus programs flag new unsigned apps. This is a false positive.
- The code is mine and the ui is claude|
If you find it useful, a ❤️ or a comment helps a lot, and feedback and feature ideas are welcome!
Description
v3 is now suporting direct refmod maker (comfy api)
without any custom nodes
FAQ
Comments (7)
Waiting for linux version
i dont have linux
but my friend has one
maybe in v4
The part that creates the dataset works great, but every time I press the Extract RefMod button, this is the error I get:
RefMod failed: Prompt outputs failed validation
Required input is missing: include_images
H3RefModCreateFromFolder: Required input is missing include_imagesWhen I run the H3RefModCreateFromFolder node manually in Comfy with the same folder input, it works, so I'm not sure why it fails from the app's end.
did you download the refmod orginal node?
@to_MacDonalds_and_beyond Yes, like I said, I can run the node in Comfy just fine.
@Derpression ah i see you are using the new version of the picker
i need to update the engine
@to_MacDonalds_and_beyond Oh ooops! I guess it must have updated today lol just my luck