How OpenAI's early gradients were made: heightmaps, layered SVG radials, and DALL·E 2

Methods for Random Gradients

How OpenAI's early gradients were made: heightmaps, layered SVG radials, and DALL·E 2

Justin Jay Wang, who worked at OpenAI from 2018, shares the techniques he used to create random gradient images for the company's visual identity. He covers three methods: heightmaps with Perlin noise and randomized color scales, layered SVG radial gradients that powered openai.com's dynamic backgrounds, and AI-generated gradients using DALL·E 2. The post includes many examples and practical prompts for generating dreamlike, floral, and sky-inspired gradients.

While the process of using them can be unpredictable — more akin to a slot machine than exact science — they offer exciting opportunities, and the technology and tools are only getting better.
  1. italiansolider

    i dont get it. Isn't that just the basic and old GLSL?

  2. oe

    Creating the gradients first with clever hacks and then with AI prompts made me a bit sad. The end results are beautiful but I feel something was lost.

  3. Ringz

    Are the AI generated backgrounds as lightweight as the first two? If not, I prefer the lighter version over the generated ones.

More from this day

2026-09-07