OpenAI Launches ChatGPT Images 2.0 With Thinking Capabilities and Better Text Rendering

[MacRumors]

OpenAI this week introduced ChatGPT Images 2.0, which the company says brings a new era of image generation. Images 2.0 is an updated model that can better handle complex visual tasks.

It is able to follow detailed instructions, placing and relating objects accurately, preserving fine detail, and rendering dense layouts. Images 2.0 is OpenAI’s first image model with thinking capabilities, and it has an improved sense of composition and visual taste, which OpenAI says will result in images that feel less AI-generated.

Images 2.0 is able to search the web to get real-time information, create up to eight images from a single prompt, and double-check its output. Graphics can be created across several aspect ratios and at up to 2K resolution. The new model also has improved multilingual understanding and can better render non-Latin text like Japanese, Korean, Chinese, Hindi, and Bengali.

Images 2.0 is available now for all ChatGPT, Codex, and API users.

Article Link: OpenAI Launches ChatGPT Images 2.0 With Thinking Capabilities and Better Text Rendering

 

[dbc34]

Thinking capabilities? We are inching closer to Skynet while everyone is creating their cool pictures to post to social media…

 

[jsavvy]

“Images 2.0 is able to search the web and pass other artist’s work off as its own.”

 

[BobRoss]

Can’t wait to create one prompt and get told I’ve hit the rate limit ❤️

 

[robprins]

just another thing we don't need 'a new era of image generation' - we so need regulation on these companies and their new slop models, they are just allowed to release stuff like this and see what happens, duh, bad things will happen, people will be fooled, it needs to stop.

 

[BeatsByTim]

Just what we need: more ai slop

 

[Thomasasz]

just another thing we don't need 'a new era of image generation' - we so need regulation on these companies and their new slop models, they are just allowed to release stuff like this and see what happens, duh, bad things will happen, people will be fooled, it needs to stop.

Fully agree. So far this was useful to big companies to lay off thousands of people and to spread infinite amount more of fake information to fool people. I do not see anymore how this contributes to public good.

 

[BeatsByTim]

just another thing we don't need 'a new era of image generation' - we so need regulation on these companies and their new slop models, they are just allowed to release stuff like this and see what happens, duh, bad things will happen, people will be fooled, it needs to stop.

it’s used for scamming old people, stealing art, and making csam. Sam Altman should be in prison

 

[Apple Knowledge Navigator]

Images 2.0 is OpenAI’s first image model with thinking capabilities, and it has an improved sense of composition and visual taste, which OpenAI says will result in images that feel less AI-generated.

And that's exactly the problem, it's entirely subjective in many instances. A computer doesn't 'think', it just runs programs that are designed to assimilate data and present in human language. The idea that this model is better because it can think more is just marketing speak for greater training.

 

[Astuces iOS]

More AI slop !!

 

[PBG4 Dude]

Wonder if this AI can de-cringify this image:

 

[neuropsychguy]

“Images 2.0 is able to search the web and pass other artist’s work off as its own.”

It really is becoming too human! 😆

 

[Capeto]

I don’t understand the hate. I, for one, am happy that every computer, phone, tablet, and gadget is exponentially more expensive just so that we can make the technology used to scam people and make illegal content even more realistic.

 

[LeeW]

Whatever you think, it's actually pretty good, big leap.

 

[symphony3299]

Wonder if this AI can de-cringify this image:
View attachment 2624076

I tried and it gave me this

 

[smkbrvo]

The more someone uses Ai, the less human they become. They simply become a bus for the information the Ai scrapes. At that point, why does that person need to be there at all?

 

[Jumpthesnark]

Images 2.0 is able to search the web to get real-time information

Translation: Now with even better copyright theft!

 

[Jumpthesnark]

I tried and it gave me this

At least they're not minors, like Grok provides.

 

[BeatsByTim]

At least they're not minors, like Grok provides.

lol if you think this isn’t designed to do the same thing

 

[BeatsByTim]

I tried and it gave me this

Now make it say “age of innocence” and make it the cover of a u2 album forced upon everybody’s phone that auto plays when connected to the car

 

[kalsta]

I tried and it gave me this

A very profound and appropriate use of the technology. I can see humanity losing more than our clothes as AI starts taking over.

 

[Danilo - Italy]

Without the API, it doesn't even reach Full HD. Plus, it takes more than twice as long to render an image compared to Nano Banana. A huge disappointment.

 

[mklnz]

Oh boy here comes the marketing hype again. There are no Al models that do any "thinking" they are just really good probability prediction models sighhh

 

[CarletonTorpin]

My favorite 'stress test' of an image-generator model is to ask it to make a skee-ball table.

Midjourney was unable to a proper skeeball table until around v6, and it took until Nano Banana 1 for Google to achieve it.

Today, skeeball tables are just 'table stakes' for an image generator, so I've had to add additional specificity to my test prompts, in order to really push the image-generator's ability.

Without commentary on the company, OpenAi's newest model has done the best yet, relative to its GPT-Image-2 image-generator.

Here's its resulting output, from the prompt I gave:

"Skeeball table on the lunar surface with an American-collonial pioneer with buckle-shoes and hat on, and a holographic coach is standing near them to cheer the person on as they roll their ball up the skeeball table ramp and make it into the top-right corner "100" circle. They have two balls remaining to roll, and they have about 12 tickets coming out of the achine so far, from their existing play."

Below is a link to this image's 'conversation' on Poe, for further experimentation (without issues most people experience when they try to make images within ChatGPT itself, whereas Poe provides API access to the image generator, so the results are better, and mode controllable).

Skeeball table on the lunar surface with an american-collonial pinoeer with buckle-shoes and hat on, and a holographic coach is standing near them to cheer the person on as they roll their ball up the

GPT-Image-2: ![Skeeball table on the lunar surface with an americancollonial pinoeer with buckleshoes and hat on, and a holographic coach is standing near them to cheer the person on as they roll thei

poe.com

 

[dss33k]

just another thing we don't need 'a new era of image generation' - we so need regulation on these companies and their new slop models, they are just allowed to release stuff like this and see what happens, duh, bad things will happen, people will be fooled, it needs to stop.

Yeah. Now Indian uncles can spread communal misinformation in Hindi too.