3 ms·
The diffusion process is conditioned on CLIP text, which works better (in theory) since the encoded text is aligned with images.
by minimaxir 2y ago
The diffusion process is conditioned on CLIP text, which works better (in theory) since the encoded text is aligned with images.