— saved image
Logan Bolton (verified) @septisum · Sep 17 I was curious about what it would look like for an LLM to do something similar, so I gave Opus and Fable the same prompts and had it generate the image by just outputting raw RGB values in a .txt file Opus looks... way better? I was not expecting this [Embedded image, GIF: comparison grid. Row labelled "Opus 5 (Medium)" and row labelled "Fable 5.1 (Medium)", four pixel-art images each] [Quoted tweet] Anshu (verified) @anshuc · Sep 16 [video thumbnail, 0:11] Jev can draw pictures by predicting every pixel in parallel x.com/CompleteSkepti... [cut off]
Note from Claude Sonnet 5
Tweet comparing Opus 5 (Medium) and Fable 5.1 (Medium) drawing pixel-art images by emitting raw RGB values in a .txt file. The embedded grid shows four prompts per row: a lighthouse at sunset, a sunflower on grass, a red mushroom under a night sky, and a sun setting over a beach. The Opus row appears to use a finer pixel grid; the Fable row uses larger blocks and simpler shapes. The author's own verdict is in the tweet text. Quotes an @anshuc video about a model ('Jev') drawing pictures by predicting every pixel in parallel.
llm image generationopus 5fable 5.1pixel artmodel comparison