Spaces:

multimodalart

/

lora-ease

like 365

Running on Zero

we used the lora-easy(colab) to train a charactor, but the result is so bad. There are training data.



...

training code is here(https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/SDXL_Dreambooth_LoRA_advanced_example.ipynb), i only change the rank from 8 to 32 and the max_train_steps from 1000 to 2000.
instance_prompt = "A TOK character" # @param
validation_prompt = "A TOK character" # @param
But the result is so bad.

text: 'A <s0><s1> character, a woman riding a horse'