Reminds me of PixArt-α which was also trained on the similarly tiny budget ($28,000). [0] How good is their result, though? Training a toy model is one thing, m

Reminds me of PixArt-α which was also trained on the similarly tiny budget ($28,000). [0] How good is their result, though? Training a toy model is one thing, making something usable (let alone competitive) is another. Edit: they do have comparisons in the paper, and PixArt-α seems to be... more coherent? [0] https://pixart-alpha.github.io/ https://pixart-alpha.github.io/