Custom animation/pixarish style LoRA created using 300 of my favorite generated still images from overnight rendering the past 2+yrs (using custom trained LoRA off other generations, etc etc) + 40 videos generated over the past few months using early attempts at making a LoRA like this, Kling earlier on, and LTXV image2video. (33frames each)
Trigger w/ "dreamyvibes style", or "in dreamyvibes style".
Description
FAQ
Comments (2)
Amazing stuff. It would be really cool for you to share some insights on training a style Lora for Hunyuna. I am bit confused with what kind of captioning I should do for the model to just pickup the style rather than the actual content of the dataset. Thanks!
Honestly I prefer just simple Dreambooth style token/class captioning - that's how this was trained. The model is pretty great at picking things up and I think the way more important thing is dataset and image to video balance mainly to keep the motion. I've had luck using HQ clips from prior training runs of the same style (that didn't quite hit the mark) to help make sequential runs get closer to what I'm hoping for. If you use discord at all search for a server called Banodoco (if you're not already in it) - it's a fantastic server w/ lots of chat activity from people working w/ the latest models (HYV being the most popular of late). Best advice I can give. And yea, thanks so much for the comment - I thought this came out great but it got buried a bit as Civitai had an issue for a few days where downloads/ratings weren't tracked - so I don't think this got presented to too many people.
Hope all is well!!
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.