Build 20 Released (Source Code & Binaries)

Build 20 Changes

Change image dimensions by increments of 8 instead of 64.

Added aesthetic scoring for saved images. Doesn't add any time at all to generation time.

Added a better quality but slower pixelization alternative

Images/prompts are now searched from an indexed SQL table. Runs at least 25000% faster but likely much faster. Slow searches from having tons of saved images shouldn't happen now.

Removed purge options from menu. Deleting image folders is now done by right-clicking the open folder buttons.

Added aesthetic score filtering for searches

Added purge function to search window to mass delete search results

Added AudioLDM

Added a music caption dataset for random music prompts (Riffusion and AudioLDM)

Added a sound caption dataset for random sound prompts (Riffusion and AudioLDM)

Fixed a bug that was causing long prompts to be truncated in png info

Added upscaling and ESRGAN model window

Added masking for the upscaling/ESRGAN window

Riffusion: Added function to convert full mp3/wav tracks to a Riffusion Project with all segment spectrograms and audio imported into the timeline

Riffusion: Added AutoGen and AutoBlend functions for imported tracks. AutoGen runs Img2Img sequentially on all spectrograms in the timeline using a newline delimited list of prompts. AutoBlend blends seams sequentially on all spectrogram pairs using a newline delimited list of prompts.