Uses existing CLiP model with an additional small pretrained model to calculate perceived aesthetic score of an image
Enable or disable via Settings
-> Aesthetic scorer
This is an "invisible" extension, it runs in the background before any image save and
appends score
as PNG info section and/or EXIF comments field
- Configuration via Settings → Aesthetic scorer
- Extension obeys existing Move VAE and CLiP to RAM settings
- Models will be auto-downloaded upon first usage (small)
- Score values are
0..10
- Supports both
CLiP-ViT-L/14
andCLiP-ViT-B/16
This extension uses different method than Aesthetic Image Scorer extension which:
- Uses modified SD Chad scorer implementation
- Windows-only!
- Executes as to replace
image.save
so limited compatibity with other non-txt2img use-cases
As a utility for batch processing, this extension can be used from CLI as well
Input param provided can be image, list of images, wildcards or folder
Score is only output to console and does not modify original file
python aesthetic-scorer-cli.py ~/generative/Demo/*.jpg
Loading CLiP model: ViT-L/14
Loading Aesthetic Score model: sac_public_2022_06_29_vit_l_14_linear.pth
Aesthetic score: 5.0 for image /home/vlado/generative/Demo/abby.jpg
Aesthetic score: 4.18 for image /home/vlado/generative/Demo/ana.jpg
Aesthetic score: 4.12 for image /home/vlado/generative/Demo/dreamkelly.jpg
- Based on: simulacra-aesthetic-models
- Training data set: simulacra-aesthetic-captions