Retrained with hand-written captions instead of auto-generating, much better and more consistent results. Start your prompt with:
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, coolky0uAnd for negatives:
source_pony, source_western, source_furryDescription
Retrained with hand-written captions instead of auto-generating, much better and more consistent results. Start your prompt with:
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, coolky0uAnd for negatives:
source_pony, source_western, source_furryFAQ
Comments (2)
Can you tell me how to make these models like these? I want the method. I want to do the same very much
Sure! As a warning it did cost money to do and is fairly technical.
1. Make a folder of images resized to SDXL resolution with the style/concept/character you want, name them numerically (1-75 in this case) and add a .txt file for each image in the folder with your tags. It can only register up to 77 tokens (each token is around 4 letters) so don’t make them too long. If it’s for pony, include the score tags to give it more influence, and come up with a unique trigger word (coolky0u here) you include right after the score tags.
2. Sign up for the service Runpod and rent an Nvidia pc, use fast stable diffusion as the starting point. Costs around $1 hour for a decent one and will take around 1 hour per 30 images in your training set.
3. Run through the SDXL Lora notebook, pointing it to pony for the base model
4. Profit
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.