2 Comments

Insightful comparison. One thing I find interesting about Gemini in the first image generation example is that they seemed to go the extra mile and make the model as unbiased as possible. I haven't seen any other diffusion model that doesn't fall in the cliche of white male programmers. I can't imagine they're doing it without injecting something into the prompt.

Expand full comment

That drew my attention too. It's certainly better than DALL-E 3 and Midjourney in this aspect.

Expand full comment