Stable Diffusion, a DALL-E 2-like system now available to 1K+ researchers, is raising tricky ethical questions by allowing depictions of public figures and more
DALL-E 2, OpenAI's powerful text-to-image AI system, can create photos in the style of cartoonists, 19th century daguerreotypists, stop-motion animators and more.
Context & Ripple Effects
Stability AI's Stable Diffusion is a DALL-E 2-class text-to-image model, but its rollout path is the inverse of OpenAI's: rather than a gated waitlist, it is landing in the hands of 1,000+ researchers at once — and unlike DALL-E 2's cartoonist- and daguerreotypist-style outputs, it will depict public figures, which is where the ethical friction starts. The access decision is the story: whoever holds the weights decides what the model can and cannot render.
Two weeks later the gate came off entirely — [[a:982249|Stable Diffusion went open source, runnable on high-end consumer GPUs under a CreativeML Open RAIL-M license that only "strives for" responsible use]] — and investors piled in, with Coatue and Lightspeed circling a round that valued Stability AI between $500M and $1B before the model line grew into SDXL and the Stable Diffusion 3 API.
First-order effects
- Over 1,000 researchers can now generate photoreal-style depictions of public figures that OpenAI's DALL-E 2 blocks, putting the burden of what-not-to-render on each researcher rather than on a central policy team.
Second-order effects
- The open-source release removes even the researcher gatekeeping: anyone with a consumer GPU can run the model, leaving the permissive RAIL license as the only governance layer between the weights and misuse at scale.
Third-order effects
- The field is splitting into two governance regimes — OpenAI's closed-access DALL-E 2 versus Stability's open weights — forcing the industry to decide whether text-to-image safety lives in access control, license text, or downstream enforcement, with funding valuations rewarding whichever side wins distribution.
The trend: Text-to-image AI is diverging into a closed-access camp and an open-weights camp, and Stable Diffusion's public-figure capability is the first stress test of whether license text can substitute for access control.