For more workflows and updates:
https://hearmemanai.com/
You can deploy the workflows in this pack using Runpod with my template:
👉 https://get.runpod.io/minimax-template
👉 Deployment tutorial
The workflow pack includes:
MiniMax – RefMod: Generate MiniMax H3 videos using a saved RefMod for consistent characters and reference details.
RefMod Studio – Images: Create a reusable character RefMod from 3–5 reference photos without training a LoRA.
RefMod Studio – Video: Capture appearance and motion from a short video and save it as a reusable RefMod.
RefMod Studio – Audio: Turn speech, music, ambience, or sound effects into a reusable audio RefMod.
RefMods Tutorial - https://civarchive.com/articles/35224/minimax-h3-refmods-reusable-identities-without-training-workflows-and-tutorial
RefMods are tiny .safetensors adapters that can be used to create reusable character identities and voices, reusable video segments and styles.
It can take images, videos or audio segments and turn into "Mini LoRAs" that can be used in Minimax workflows.
Description
FAQ
Comments (9)
thanks for posting my node!
I tried this node and could not figure out how to use it lol. I've been so stoked to try it, thank you both of you for your hard work!!
Thank you so much for the great work Luisa_caotic. Any chance we can make it work with Krea2?
Thank you for your work, much appreciated
this works really well. A custom audio node would be awesome
Works great, thank you. Is there a workflow for using two or three models at the same time?
Hey hey great workflow its working great , took a bit of testing but I m now running it with sage attention and turbo 6 steps which dropped the generation time by a lot , only issues I have is that when i try multiple character the faces works and are different but the body is the same for both subject even though I was using different body ref when creating the 2 refmod people, is there a way to make sure they keep their own bodies ? I tried so many different prompts and strength but nothing seems to work
i love the concept, unfortunately the prompting is extremely confusing. I'm finding my videos are constantly getting characters bleeding into each other as far as their characteristics. I dont understand how the prompts are supposed to identify and differentiate each refmod. I understand we're supposed to define them in the <Subject 1> <Subject 2> etc...fields, but I'm finding that, if I'm using two or more characters that are remotely similar in appearance, the model seems to have a hard time understanding which refmod to use and the characters often get mixed up. I'm finding this happening much less often using the standard <Picture 1>, <Picture 2> <VIdeo 1> etc.../standard ref2va prompting route. :( Any suggestions?
in your subject definitions, try describing each subject more uniquely. even if the characters are similar I've found the distinction between character descriptions help keep them separate. and of course avoid prompting the scene with "she" or "her" etc. always use S1/S2 or <subject 1>/<subject 2>