2 ms·
PixArt-α: Fast Training of Diffusion Transformer for Text-to-Image Synthetis
- Palmik 3y agoThis is a really impressive result -- a model that rivals (and I would say is better than) SDXL that cost only low five figures to train. It is really good at following complex prompts, you can find some examples here https://www.youtube.com/watch?v=ZiUXf_idIR4 https://www.youtube.com/watch?v=ZiUXf_idIR4 at the 16 minute mark.