Livestream: 2026-04-21T19:00:00Z
Join the stream via YouTube:
Gabriel Goh, Kenji Hata, Kiwhan Song, Alex Yu, Boyuan Chen, and Nithanth Kudige join host Sam Altman to introduce and demo ChatGPT Images 2.
Comments on YouTube are disabled. Join the conversation here in the community and share your impressions, questions, and insights.
It’s been a while since the last livestream. I’m really looking forward to it.
What are we going to see?
I am guessing, of course could be off by a long shot, but…a new image generation model that can accurately replicate/generate UI and text?
My guess is also on the new image generator.
Yeah, that’s my quess too. Hoping that there will be an API for it in addition to availability with ChatGPT…
Wouldn’t be surprised if they did this.
Oh yes, I remember seeing rumors about an image model.
I also recall something about GPT-6 being available to everyone. But I do not want to build up expectations because I really do not know what is going to happen.
My real test will be spritesheet creation. So far I’ve been able to use Gemini/Nano Banana Pro to great impact, with some additional (considerable 
) editing in aseprite, but in general it’s been great for both sprites and tilemaps. (as a funny aside, Gemini seems to struggle with the walking sequence).
The next frontier model will come. Maybe too much for a livesteam to handle the details of everything 
The stream is live for those who want to wait.
We are live now!
Imagegen 2.0
Then what I read today was true​
A sample with new model in ChatGPT:
A massive flock of starlings performing a dynamic murmuration across a wide open sky, fluid organic shapes swirling and folding in midair, captured in a wide-angle composition with the flock centered against a bright midday sky, clear blue background with soft white clouds, high-key lighting with sunlight illuminating each bird, lifted shadows and even exposure, vibrant natural tones with subtle contrast between dark birds and bright sky, tack-sharp detail capturing individual wings in motion, crisp edges, no motion blur, clear air, no haze or vignette, high-key lighting, exposure raised (+0.7 EV feel), lifted shadows, daylight white balance (6500K), vibrant saturated colors, ultra-clear detail, ultra wide size, —ar 21:9.
This new generator should be better at generating more realistic images​ Now I really need to study more about imagegen 2.0.
^.^ next model they’ll write an entire encyclopedia on a spec of dust.
Looks like fun
I did some spritesheet tests and it’s on same-or-slightly-worse level compared to Gemini. Previously I would get very “mangled” 8-bit sprites, now they are at least showing properly (used extended thinking here).
Probably optimised more for the use cases in the demo like realistic images and creating comics and infographics.
Can’t find the link to the API doc yet. If anybody know, please post
