OpenAI releases DALL-E 2, a higher-resolution and lower-latency version of its text-to-image generation program with new capabilities, to researchers in preview
Researchers can sign up to preview it — Artificial intelligence research group OpenAI has created a new version of DALL-E, its text-to-image generation program.
The VergeAdi Robertson
Context & Ripple Effects
DALL-E 2 begins as a controlled research preview, establishing OpenAI's staged-access approach before the model moved into a developer API public beta and a credit-priced wider beta. The release also placed OpenAI in a nascent competitive set that soon included Google's Imagen.
First-order effects
Researchers gain early access to a faster, higher-resolution DALL-E generation system, while OpenAI keeps distribution limited to preview participants.
OpenAI turns text-to-image generation from an earlier model demonstration into a product candidate whose performance and capabilities can be tested by external researchers.
Second-order effects
Google Research's subsequent Imagen disclosure positioned a direct rival against DALL-E 2, while declining to offer public code or a demo.
OpenAI's limited preview created the basis for later commercialization through a per-image developer API and a credit-based beta, extending access from researchers to builders and users.
Third-order effects
The DALL-E line points toward generative-image systems becoming features inside broader AI interfaces: DALL-E 3 was later controlled through ChatGPT and offered to paid and enterprise subscribers.
If staged previews, safety preparation, and paid distribution remain paired, access governance will become part of how frontier image models are commercialized rather than a temporary launch constraint.
The trend: Generative-image models are moving from restricted research previews toward governed, paid distribution through developer platforms and general-purpose AI assistants.
Our newest system DALL·E 2 can create realistic images and art from a description in natural language. See it here: https://openai.com/... https://twitter.com/...
DALL·E 2 is here! It can generate images from text, like “teddy bears working on new AI research on the moon in the 1980s”. It's so fun, and sometimes beautiful. https://openai.com/... https://twitter.com/...
Artificial intelligence just took a huge leap forward. You can now type literally any phrase such as “a koala dunking a basketball” and OpenAI's DALL•E 2 system will create a realistic image https://twitter.com/...
Incredible examples of computer-generated art in the replies to the tweet below. Most of these would pass an “Art Turing Test” for me: I'd rather have these computer-generated artworks on my wall than a lot of human-generated art. https://twitter.com/... https://twitter.com/...
This is mindblowingly neat. That it gets so styles well isn't the big thing, it that it actually gets compositionality so much better. This is not just teddy bear parts in the vicinity of each other but forming reasonable full bodies. The examples in the blog post are fun. https:…
@Techmeme @thedextriarchy this super impressive, as it can fit formats well too https://twitter.com/.... it's fun to play with and think of the possibilities. makes you wonder about IP infringement/attribution issues, nonetheless it will be hugely disruptive to graphic designers …
its attempts at pixel art are pretty impressive, here's a sample of images generating from the prompt “An astronaut lounging in a tropical resort in space as pixel art” https://twitter.com/...
Hierarchical Text-Conditional Image Generation with CLIP Latents paper: https://t.co/cS10wW0ouZ propose a two-stage model: a prior that generates a CLIP image embedding given a text caption, and a decoder that generates an image conditioned on the image embedding https://t.co/prQ…
DALL·E 2 — generate any image from a text description. Imagination is the limit. “A Shiba Inu dog wearing a beret and black turtleneck” “A photo of a quaint flower shop storefront with a pastel green and clean white facade and open door and big window” https://openai.com/... http…