See how we can manipulate generated images by adjusting SAE features trained on various transformer blocks.
Let’s try to generate an image with an empty prompt and enable only one feature. This results in meaningful images highlighting the same concepts as above: faces, dishes, lights and tents!
Take a look at images where these features are most prominent. They correspond to similar objects as above. E.g. 4539 activates on funny animal faces, while 450 highlights dishes.
1
6
441
First, we generate an image with a fun prompt. Below are the SAE features that are most active during a forward pass through one of transformer blocks.
1
3
458
Excited to share our latest breakthrough! We trained sparse autoencoders to decompose intermediate results of SDXL Turbo's forward pass. These autoencoders learn highly interpretable features that can be used to manipulate the image generation process.
arxiv.org/abs/2410.22366



