IG/TikTok Aesthetic LoRA for Minimax H3
Authentic Instagram & TikTok aesthetics
Trained on T2V but supports also I2V
720x1280 resolution with 15-30 steps
Weight 0.4-0.8 is recommended
⚠️ Avoid prompt bleeding using shorter sentences or words
Description
First initial release for Minimax H3.
FAQ
Comments (12)
looks promising! i will try it soon!
ok i tried it out! its pretty good! i have a lora request if you are ok with that! could you do a gyrating hips lora specifically on IG and tik tok girls? basically, i think a gyrating hips or "hip roll" version of this lora would be awesome!
Does this work with ref2V? any chance a finetune on that?
If i understand correctly you need to train LoRA separately for that, but i will check it out!
It does "work", see my example. But the difference to the same prompt without the lora is very small. I think the ref model mainly using the input pics for the "style".
@Lora_Addict You are using 8 steps, when you take away the quality it doesn't always help you that much.
@AdaptiveVision Yeah, using turbo lora with 8 steps. I tried it without and 20 steps too. As i said, you can see the lora working, the difference is just very subtle :)
super nice work man, t2v is great. havent figured out i2v with this lora.
I just blew 20 hours on an rtx6000 training a lora on 600 youtube clips of UGC 1080p. Afterwards, no goddamn difference at all at any strength.
@bambushu Its not about how much data you have, you also need to know how to curate and edit the data.
hi did u used images or videos in your dataset?
I been always using videos, but im planning to test images as well.