Timeline
Post
Remote status
Context
5
@sun unrelated but idk wtf are ai gemerators doing now. They don't seem to be purely scalaed data as diffusion alwaya makes text messy.
But in this caae it doesnt even seem to be text rendering otjer models use because the typography is genuinely quite bad
But in this caae it doesnt even seem to be text rendering otjer models use because the typography is genuinely quite bad
@meeper I can't quite tell the difference between bad human typography and bad ai typography
@sun I mean typography should just be the font engine
Whats uncanny is the text has zero kerning (if it was what it is called) - They have equal spacing between glyphs, so maybe its jist genrating a chatacter + coordinate kinda like a pdf, clearly these dovs are made woth symbolic stuff
Whats uncanny is the text has zero kerning (if it was what it is called) - They have equal spacing between glyphs, so maybe its jist genrating a chatacter + coordinate kinda like a pdf, clearly these dovs are made woth symbolic stuff
@meeper Claude seems to do better if you ask it to make an SVG than ask it to make an image with text
text rendering in latent noise diffusion models is always going to be a limitation, Open AI adds additional post processing to render the text in their image generation.
Replies
0
No replies yet.