Google Research details Imagen, an AI-based text-to-image generator to rival OpenAI's DALL-E 2, and says it won't release code or a public demo at this time
The AI world is still figuring out how to deal with the amazing show of prowess that is DALL-E 2's ability to draw/paint/imagine … Source: Google Research .
TechCrunchDevin Coldewey
Context & Ripple Effects
Google's Imagen announcement is a direct response to OpenAI's unveiling of DALL-E 2, which reached researchers in preview a month earlier with higher resolution and lower latency. Google is claiming rival capability while choosing a stricter access posture: no code, no public demo.
OpenAI's DALL-E 2 loses its uncontested researcher-preview window — Google is now a named peer in text-to-image, and the comparison between the two systems becomes the benchmark researchers and press will chase.
Google Research keeps Imagen out of public hands for now, so outside developers and artists cannot test it directly, leaving Google's quality claims unverifiable except through its own paper.
Second-order effects
Access policy becomes a competitive lever: OpenAI's staged researcher preview and Google's tighter gate mean each lab's release cadence, not just model quality, shapes who gets mindshare in generative imagery.
Google's own pipeline is the knock-on effect — the gated model gets funneled into controlled channels like AI Test Kitchen, giving Google a way to gather usage data and safety signals before any broader product launch.
Third-order effects
If the pattern holds, text-to-image models stop being research artifacts and become product infrastructure: Imagen's path from withheld demo to Imagen 2 powering Bard, Vertex AI, and ImageFX shows labs converting gated models into consumer and enterprise surfaces.
Staged, safety-gated access hardens into the industry template for releasing generative media models — labs control exposure through invite apps and previews first, which also concentrates early commercialization inside a few large platforms.
The trend: Text-to-image generation is shifting from gated research previews toward embedded commercial products, with each lab's access policy doubling as its competitive strategy.
AI can unlock joint human/computer creativity! Imagen is one direction we are pursuing: https://gweb-research-imagen.appspot.co m/ “A high contrast portrait of a very happy fuzzy panda dressed as a chef in a high end kitchen making dough. There is a painting of flowers on the wal…
Google just announced a DALLE-2-like model: Imagen For now no code, just demo site: https://gweb-research-imagen.appspot.co m/ And paper: https://gweb-research-imagen.appspot.co m/ ... https://twitter.com/...
Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding project page: https://gweb-research-imagen.appspot.co m/ sota FID(7.27 on COCO), without ever training on COCO, human raters find Imagen samples to be on par with the COCO data itself in image-text ali…
Imagen also struggles with generating photorealistic images of people, though perhaps that's a blessing for now. You can expect these capabilities to become public in the near future, and imagine how systems like this could be used for hoaxes and harassment.
Caveats apply, though. Google is not releasing the model publicly for safety reasons, so the images here have been hand-picked. Its own evaluations also found the system had serious biases, “including an overall bias towards generating images of people with lighter skin tones.”
Google's has unveiled its new text-to-image AI system, Imagen. Type in any text you like and the system generates a corresponding image. It's extremely expressive stuff — and Google says produces higher-quality results than OpenAI's rival DALL-E. https://www.theverge.com/... http…
A state-of-the-art text-to-image generation model (like DALL-E) from Google Brain! 🚀 Imagen - unprecedented photorealism × deep level of language understanding 🔥 Website: https://gweb-research-imagen.appspot.co m/ Paper: https://gweb-research-imagen.appspot.co m/ ... https://twit…
any idea why Google Brain is publishing the result of their latest big research project in an appspot url that doesn't even have the string ‘google’ in it? https://twitter.com/...
Google's answer to OpenAI's Dall-E is “Imagen” and it looks waaay more refined. There's no denying this will be a problem for illustrators and creative professionals in the future. https://twitter.com/...
Looks like Google's Imagen might produce even better results than OpenAI's DALL-E. Shame Imagen won't be available for now because it is trained on nsfw data. https://twitter.com/...
More remarkable AI. I've no fear this work will replace true art. But I do worry it might replace some of the pay-the-bills work artists do to fund their creativity elsewhere. https://twitter.com/...
We are in a generative art race, with new and better engines every week. Here is Google's latest, which is very good. These really will change art. Everyone will art. https://gweb-research-imagen.appspot.co m/
Looking at Google's new image-generator (better than DALL-E, they argue), I'm pretty sure there will soon be a good use case for chatbots-as-interface-to-the-visual- realm: https://gweb-research-imagen.appspot.co m/ https://twitter.com/...