09-20-2026 - Add Starfire Lora with proper audio
I'm still trying to learn the right way to clip videos and caption for good audio training. My plan is to make Starfire and then the next release will be Starfire and Raven in the same lora. If I can, then I'll keep adding characters with each release.
First Lora style works great but has no audio training.
09-17-2026 - Initial release
I tried to train these characters into the style lora:
Starfire - strfr, red hair, long hair, green eyes, orange skin, crop top, midriff, miniskirt, thighhighs, thigh boots,
Raven - rvn, purple hair, short hair, purple eyes, grey skin, forehead jewel,
Jinx - jnx, jcolored skin, grey skin, hair horns, pink eyes, pink hair, choker, long sleeves, pantyhose, striped clothes, striped pantyhose, black dress, wide sleeves, collarbone,
Robin - rbn, black hair, red shirt, green sleeves, short sleeves, mask, superhero, domino mask,
Description
strfr, long straight red hair and green eyes,
purple crop top, purple skirt, silver shoulder collar and armbands,
<lora:StarfireTT_H3_v1:1.0>,
FAQ
Comments (12)
Damn... I'm surprised your doing minimax loras. I'm assuming you using both images and videos in this lora? Could describe how you went about captioning it. Did you use booru captioning for the images and neutral language for video clips.
YEEEEEEEEEEEEEEEEEEEEEEEEEEEEESSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSSU
I love that she is drinking from mustard. It's something she'd do, then say human drinks are delicious.
Oh my god.....so they can sound just like the characters too. I think i'm going to be busy again...
Lol IKR!
Its super exciting that we can make characters with accurate voices but on the other hand its a huge headache to make a dataset with voices lol.
I've built some tools to help split up videos into good sizes clips for training with automatic transcripts but the automation is a little bit messy so these datasets are extremely manual.
@CitronLegacy Oh yeah? You have an article?
@Geno I have have an article on the training now https://civitai.red/articles/36097/training-h3-loras-on-small-datasets-what-ive-learned-so-far
Also the dataset for Starfire is available in the article.
@CitronLegacy Thank you! I was theorizing prior to you responding that it would be possible to have a dataset of both images and videos. And you now confirmed it. Then i can begin my tests with a character. I have the perfect formula for getting accurate model sheets for a single cartoon character, as well as making videos with said character. Both can help out this dataset greatly. I do have to ask because im still new to minimax 3, how do you load the lora node properly in comfy for the minimax template if you could?
@Geno Nice glad to hear you have a plan!
I think its hard to get good videos for training.
Here is my simplified workflow for loading loras https://github.com/citronlegacy/comfyui_workflows/blob/fbc1ffb39f6616dd5aa2ca69dfc094087f0005df/video/minimax_h3/MiniMax_H3_Workflow_v001.json
You just have to list the loras in the "Load LoRA Tag" node. They wont be part of the prompt but they do get loaded for generation.
If you don't want to use this node you can just chain multiple "Load LoRA" nodes for each lora. I included a Load lora node for the turbo Lora which I usually use at weight 1.2 which allows for generating videos with only 6 steps. This workspace is optimized for my RTX 5070 so you may need to adjust things if you have a better or worse GPU.
@CitronLegacy Try making the character as references in grok imagine. If a character's screentime is low, ascertaining the body for a model sheet, and then using it to make a video can work too. Some of my loras were created that way, as well as some unreleased ones.
@Geno Very nice! I loved using Grok for datasets when it was free. Are you using Grok through CivitAI or through the official site? If official, what subscription did you buy?
@CitronLegacy Official site. They gave me a discount on heavy before for 3 months, so it was 99 at the time. But i did it also back when it was just 30 for higher rates and limits. Im on super grok plus-99