Cog-SDXL
This is an implementation of the SDXL as a Cog model. Cog packages machine learning models as standard containers.
Development
Follow the model pushing guide to push your own fork of SDXL to Replicate.
Basic Usage
for prediction,
cog predict -i prompt="a photo of TOK"
cog train -i input_images=@example_datasets/__data.zip -i use_face_detection_instead=True
cog run -p 5000 python -m cog.server.http
Update notes
2023-08-17
- ROI problem is fixed.
- Now BLIP caption_prefix does not interfere with BLIP captioner.
- Now lora_url can be used to accept lora models from the web.
2023-08-12
- Input types are inferred from input name extensions, or from the
input_images_filetype
argument - Preprocssing are now done with fp16, and if no mask is found, the model will use the whole image
2023-08-11
- Default to 768x768 resolution training
- Rank as argument now, default to 32
- Now uses Swin2SR
caidas/swin2SR-realworld-sr-x4-64-bsrgan-psnr
as default, and will upscale + downscale to 768x768