Early Stable Diffusion models primarily worked well on white people
Wesley Tian · 1st-time founder grows AI headshot app from $0 to $10M ARR in 2 years—with no funding. | Wesley T... · PMF Show · Apr 24, 2025 · at 23:02
Wesley Tian, founder of Aragon.ai, discusses the demographic bias and dataset limitations in early Stable Diffusion image generation models.
0:00 / 0:15exact quote · 15.6s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“It only worked for, like, I think, like, certain ethnicities, like, because the model, like, say with Fusion was, like, trained, like, it wasn't trained on a very comprehensive data set. So maybe like only working on like white people. Right. And so like for everyone else, it would just be like, you know, he's like terrible.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Wesley Tian
Tian posted fake success stories on Reddit to acquire early users
“I basically, like, went on Reddit, and I had this idea to, like, pretend I got a date by using Eragon to create photos for my dating app. So I went out to slash r slash Tinder, and I was like, Hey guys, I used Aragon to, like, create these pictures of me, and,…”
Tian predicts all AI headshot tools will reach 99% quality within a year
“I think maybe in a year, like, we'll all get to, like, sort of, like, 99%, right? And so, I do think there's a bar, and I do think brand and distribution plays a big difference.”
Wesley Tian says Canva's AI headshot product is 'terrible'
“And Canva's product is terrible, by the way. They have a free product, but it takes a few seconds, but the output doesn't even look like you.”
Aragon claimed thousands of users on its site before acquiring them
“Early on, like you know, I think a lot of companies do this, but like, we didn't have that many users, but we still put, like, used by thousands of people. And by doing that, we actually got to thousands of people.”