
researcher @openai, imagegen
San Francisco, CA
Joined October 2016
- Tweets39
- Following179
- Followers519
- Likes147
Pinned Tweet
Our new model gives users even more control to get what they need, but the model also just groks user intent **really** well!
Discovering that the model can make flawless 360 panoramas from a simple request was a huge surprise for me, and I knew I had to put this in my demos!
Dibya Bhattacharjee retweeted
🥹 i feel like my teammates are even more special than the model itself.
Dibya Bhattacharjee retweeted
just moving the imagegen bar up
More exciting news from @OpenAI: GPT-Image-2.5-Sunburst is #1 and Flare is #2 across the Text-to-Image Arena, the Image Edit Arena, and the Multi-Image Edit Arena!
Both releases are significant improvements over GPT-Image-2 (medium) in all 3 Arenas. Sunburst delivers the highest quality generation, while Flare still improves upon GPT Image 2 at significantly faster speeds.
The increase by points are (Sunburst | Flare):
- Text-to-Image: +40 | +18 pts
- Single-Image Edit: +59 | +30 pts
- Multi-Image Edit: +81 | +47 pts
Across all categories like 3D Imaging & Modeling, Art and more, Sunburst is #1 and Flare holds #2 in all three Arenas.
Congrats to the team at @OpenAI!
We've pushed the frontier yet again with GPT-Image 2.5, but personally I'm most excited about the real improvements to multi-turn editing with this release! Great stop motion animation is one of many new use cases unlocked by this capability!
2
11
814
Dibya Bhattacharjee retweeted
Images 2.5 is here.
I don't think it can solve super difficult math problems, but it is really good and we hope you enjoy it.
openai.com/index/introducing…
ChatGPT Images 2.5—faster, sharper, smarter, with better tools for creating whatever you can dream of.
- Faster image generation to keep your ideas flowing
- Improved fidelity for more natural, recognizable images
- Consistent details across multiple edits
- Comment-based edits to change only what you want
Dibya Bhattacharjee retweeted
While a resolution to Navier-Stokes is the highlight of today, this plot is really the headline result to me.
Our ongoing training run demonstrates unprecedented performance on mathematics research, as measured by our model's estimated performance on a collection of **open** math problems.
Of course, to settle Navier-Stokes, we took this model and also scaled test-time compute allll the way up.
Replying to @OpenAI
This model represents a step-function improvement on many benchmarks, and its training is ongoing.
Our internal model group arrived at the Navier–Stokes solution in 88 hours, using around 10,000 coordinating AI agents.
Throughout the effort, we maintained the strict safeguards—including monitoring and isolation—that we apply to all our frontier evaluations.
Dibya Bhattacharjee retweeted
ChatGPT Images 2.0 is helping people and businesses unlock new ways to create and build.
Compared to just a few weeks ago, we’ve seen a more than 50% increase in image usage. Nearly 60% of daily users are newly logged-in users.
Like our users, we just keep building, and we’re continuing to ship new features — including the new 360° viewer available on desktop now and mobile next week.
Dibya Bhattacharjee retweeted
You can now discover 360 worlds within ChatGPT Images on web! Just ask for a "360 image" and click "enter 360 world." Excited to see what worlds you create!
Dibya Bhattacharjee retweeted
more from gpt images 2. personal website designs for a systems engineer.
im so creatively inspired by this sh*t, man
🤖 Made with AI
okay sooo GPT-5.5/Imagen 2 generated all three of these. some of the greatest ui's i have ever seen from a model. especially considering they are one-shot
its important to note that i handed them one single screenshot of a bunch of abstract art from pinterest and asked them to use it as their inspiration, but not to copy anything from the image, and to only use it as inpiration when crafting the ui's completely from scratch.
freaking stunning
Dibya Bhattacharjee retweeted
Comparing GPT Images 2.0 to Nano Banana 2 is apples to oranges.
What you can do with it is what matters.
10 ways people are leveraging the new model:
Dibya Bhattacharjee retweeted
Can’t emphasize this enough. I was locked out of Pro image gen due to a deployment bug (fixed just tonight), and it’s much better.
The old intuition says model choice and image quality are unrelated because it all funnels to a txt2img tool call, and that intuition is now wrong.
Dibya Bhattacharjee retweeted
Headshots are a top use case on ChatGPT Images. We have the #1 model for making you look like a more professional and polished version of yourself, while actually keeping your appearance!
Dibya Bhattacharjee retweeted
this graphic is must more dramatic, i approve.
GPT-Image-2 had a 93% win rate in Image Arena.
Arena rankings come from blind, pairwise battles where voters pick between two anonymized image outputs for the same prompt.
GPT-Image-2 from @OpenAI was preferred 93% of the time, resulting in a record-breaking +242 point leap ahead of the competition on Arena's Text-to-Image leaderboard.
Dibya Bhattacharjee retweeted
yooo ok...i think Chatgpt Images 2 is the most impressive visual reasoner i have ever seen. this is first thing i have ever generated with it. first attempt.
i'm genuinely so unbelievably impressed right now.
I gave it the README for the memory system I just open-sourced (mnemos) and simply asked it to generate a blueprint explaining the contents of the markdown file.
i gave no context, no explanation of what was even in the markdown. and it absolutely nailed it. not just in terms of spelling/text rendering. every concept is accurate. every detail is true to the memory system, as far as i can tell.
I can't even try to keep it lowkey rn. this is nutty, man.
prompt: Create a single-subject blueprint poster explaining the core concept of the uploaded document. Render it as a clean white technical line drawing on a deep cobalt-blue blueprint background with a subtle grid. Use flat 2D schematic linework, orthographic contours, construction lines, measurement ticks, arrows, and minimal technical labels. The composition may also include 2-3 smaller inset studies or alternate visuals. Keep the layout simple and balanced with no shading, no 3D, no lighting, no extra objects.
(prompt adapted from original version designed for single subject content)
second video is the markdown i gave it.
🤖 Made with AI
Dibya Bhattacharjee retweeted
what. what. what.
gpt-image-2 almost passes the pelican test...in a screenshot of a code editor.
🤖 Made with AI
Dibya Bhattacharjee retweeted
"School worksheet with maze, 32 wide by 48 tall, rows and columns numbered, no unreachable rooms, solution in blue pen, name field reads 'ChatGPT Images 2.0' in cursive at top"
Note the fusion of code and imagegen this task requires.
🤖 Made with AI
Dibya Bhattacharjee retweeted
With the new ChatGPT Images 2.0, text-to-image generation finally feels like a true programming experience. I was part of the ChatGPT Pro Community early testing group. From my perspective as a research mathematician, this new model is a huge boost to imagination, allowing me, with just a few words, to sketch, draw, plan, and visualize intuitions that I previously could only keep in my head. The model is blazing fast, highly steerable, and extremely responsive to suggestions and changes. I am attaching a small gallery of recent experiments I conducted during the testing phase.
Program your thinking to make the magic happen - now imagination is the only real limit.
🤖 Made with AI
Dibya Bhattacharjee retweeted
amazing!
Using the new ChatGPT Images 2.0 model, I believe I have now created one of the most detailed posters of the immune response to viral infection ever made!
What shocks me is that I cannot find any significant errors, even after reviewing it twice!
Even more shocking and unbelievable is that the entire piece is drawn like a masterpiece and reflects incredibly deep insight, and understanding of the immune response, which is one of the most complex biological events.
I will print a poster version of this as a poster and hang it on my wall!
This model simply would not exist without the insane grind from @kenjihata. Thank you Kenji!! 🙏🏼
I still remember how stunned I was the first time I tested GPT Images 2.0 on a really wide aspect ratio on a random prompt and it just worked. I didn’t save that exact image, but here’s a recreation.
1
13
1,015
Dibya Bhattacharjee retweeted
GPT Image can nail your infographics creation with minimal prompt. Yuguang in our team spent a lot of time improving the infographics ability. Please also check out his official demo video!




















































