Alita: Battle Angel Wan 2.2 LoRA
main gallery is here: https://civarchive.com/posts/25345133
Trained in dual-mode in musubi-tuner. One LoRA file. 35 images.
Trained at 256 x 256 pixels, then 512 x 512, then 768 and 1024 as well.
Take your pick. Examine the quad comparisons. They are small files (150mb each) so you can download all four and still be under the size of some half LoRAs.
For Alita the facial likeness is solid at 256 it seems. She has a fairly human face, so the base handles her fine. At 1024 things get weird.
The "training_data.zip" includes EVERYTHING you need to reproduce my results, including the training data and configuration files as well as install instructions and custom python scripts for processing data to train your own LoRAs.
Zip is here:
https://pixeldrain.com/u/YV1zNvjF
Description
This version was trained at [512,512]. All other parameters are the same.
FAQ
Comments (4)
Interesting results! I appreciate you posting your findings and LoRAs. 256 is surprisingly good...512 looks the best to me. I didn't see the training_data.zip file, could you please share that?
Please make one from tekken game series
💕👍🏻
Thank you so much for doing these tests. 768 does feel like the sweet spot regarding speed and accuracy.
And before anyone judges the mechanized body details, know that the training data only captioned everything as "Alita", and she had multiple bodies in the movie, so if you either were training someone with a consistent body, or captioned the different bodies, you probably would get more consistent body details with less resolution than it appears here.
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.