All the intermediate checkpoints can be found in https://huggingface.co/alea31415/tenten-characters
The base model is ACertainty
Trained at clip skip 1.
For LoHa you may need to adjust weight according to the base model that you are using, and switch to an earlier checkpoint (tenten_loha_step-20000.safetensors or tenten_loha_step-30000.safetensors) may also help.
If you find the anime style too strong you can try to put aniscreen in the negative but with a small weight.
On the other hand putting aniscreen in the positive recovers the full anime style.
Example Images
LoHa: 2,3,9,10,12,13,14,15,16,18
Checkpoint: 1,4,5,6,7,8,11,17,19,20
--
I don't need support or credit, but I would be glad to know that you are using the models I trained and find it useful.
Moreover, I would like to advocate for more franchise models.
You can take a look at my workflow https://github.com/cyber-meow/anime_screenshot_pipeline if you are interested.
I just want to spread the fact that there is no reason to encode a single concept in each lora.
Description
Resumed from checkpoint v1 and trained at resolution 768, batch size 4, for another 16582 steps.
The model is more stylish and produces more detailed background, thanks to training at 768.
However the model is also quite overfitted, and getting rid of the anime style is more difficult.
In my opinion, this is the most difficult to use among the three version but it also has the greatest potential.