I hit 100 AI-generated images in one day and it changed my mind about quality vs quantity

I always thought you needed to spend HOURS tweaking prompts for a single good AI image. Then I started a side project last Tuesday where I just blasted through 100 generations in a day using Stable Diffusion. About 20 of them were actually usable and 3 were genuinely impressive. The speed let me experiment with wild ideas I never would have tried before. Has anyone else found that generating tons of images actually improved their results?

2 comments

2 Comments

betty_scott1810h ago

Haven't you found that going for quantity makes you kind of sloppy though? In my experience, when I try to blast through a ton of generations I end up relying on the same prompts and styles over and over. The 3 impressive ones you got might have come from lucky accidents rather than actual learning. I'd rather take my time on 10 really careful prompts than sort through 100 duds. Your mileage may vary, but for me the quality comes from focused effort, not just volume.

troy9968h ago

The 20 usable ones out of 100 actually taught me more than 10 careful prompts ever did. When you blast through that many, you start seeing patterns in what works and what doesn't really fast. Those 3 good ones weren't just luck - they came from me noticing halfway through that certain lighting keywords gave way better results, and I doubled down on those. The volume let me test 15 different composition ideas in an hour, and I would have never tried half of them if I was being precious about each image. Quantity gives you a bigger sample size to learn from, simple as that. You can't argue with data when 100 tries shows you something 10 tries never would.