OpenAI

No more extra fingers? The good, bad and ugly of ChatGPT Images 2.0

Tech content creator Marcin Teodoru described an image output he received from ChatGPT Images 2.0 as "pixel perfect."

USA TODAY

April 23, 2026, 3:20 p.m. ET

OpenAI launched a new image generator that may cut down the number of extra webbed fingers circling the internet.

On April 21, OpenAI launched ChatGPT Images 2.0. Released a year after the initial ChatGPT Images, the "state-of-the-art model" can search the internet for real-time information; process up to eight outputs at once; better create small text, iconography and dense compositions; offer renderings in a wider range of languages; and generate photos in varying aspect ratios, according to an OpenAI news release.

"Instead of getting something vaguely in the neighborhood of what you meant, you get something you can actually use," OpenAI said about the new generator in a news release.

ChatGPT Images 2.0 comes less than a month after OpenAI announced it was discontinuing its video generator, Sora. In December, Walt Disney Company announced it was making a $1 billion investment in OpenAI, allowing Sora users to create their own AI-generated Disney, Pixar, Marvel and Star Wars characters. The partnership has since ceased.

As more users try OpenAI's new image generator, here's what tech and media experts think.

What makes ChatGPT Images 2.0 better?

According to an OpenAI news release, here are the new features ChatGPT Images 2.0 offers:

An AI-generated promotional flyer created using OpenAI's ChatGPT Images 2.0, which launched on April 21, 2026.

"Thinking capabilities" that can search for real-time information on the internet, create up to eight images from one prompt and double check outputs
Training knowledge cutoff of December 2025, in hopes to provide the most up-to-date, accurate information
Refined small text, iconography, dense compositions and stylistic constraints
Up to 2K resolution
Stronger understanding and rendering in non-English languages, including Japanese, Korean, Chinese, Hindi and Bengali
Greater aspect ratio ranges, as wide as 3:1 and tall as 1:3
Cross-platform collaboration with OpenAI's coding agent Codex

ChatGPT Images 2.0 also offers a new "thinking model," which takes more time to process a prompt in order to create a more refined output, according to the news release. Dependent on the prompt, this process may take a few minutes, OpenAI Product Manager Adele Li told USA TODAY.

Does ChatGPT Images 2.0 actually create stronger images?

ChatGPT Images 2.0 has received mixed reactions from users, with some thoroughly impressed by the new and improved outputs, while others are disappointed.

In a TikTok video posted on April 22, AI for Life founder Celia Quillian shared a handful of images she created with the new generator, including a full-body portrait of a woman and collage testing six different hairstyles on herself. She finds the generator's new capabilities impressive.

An AI-generated image created by AI for Life founder Celia Quillian with ChatGPT Images 2.0. This image was created with the following prompt: "A vertical iPhone photo, casual and colorful, taken handheld in the bright afternoon. A woman in her late 20s — natural features, a few freckles, hair a little wild from humidity, genuinely happy — stands in front of a wall completely covered in a massive painted mural in the Roma Norte neighborhood of Mexico City. The mural behind her is vivid: hot pink, electric yellow, cobalt blue, figures of women and flowers and birds. She's wearing a burnt orange linen co-ord set, hoop earrings, white Birkenstock sandals. She's holding a michelada in one hand and doing a small shrug-smile at the camera — an "I'm in Mexico and it's noon and I have a drink…sue me~" face. The iPhone's auto HDR has made the mural colors almost unrealistically saturated while her skin looks warm and natural. Shot at roughly eye level, very slightly wide. 9:16 vertical."

"This should be blowing your mind right now," Quillian said in her video. "It should also have you concerned because when you can make something this realistic in a matter of minutes and anyone can do it for free, that's going to open up a whole other can of worms."

Tech content creator Marcin Teodoru described a ChatGPT Images 2.0 output he received as "pixel perfect." In a TikTok video posted on April 21, Teodoru asked the image generator to analyze a screenshot of a website and create a version in dark mode, a user interface that displays text on a dark or black background. The result came back clear and accurate, with the website's text and imagery the same as the screenshot Teodoru submitted.

See examples from ChatGPT Images 2.0

Woody Hood, director of critical and creative media and film and media studies at Wake Forest University, wasn't impressed after playing around with the new image generator.

Woody told USA TODAY the generator's photorealism is done well but the proportionality of objects is "odd." Playing around, Woody asked ChatGPT Images 2.0 to create an image of a chicken chasing a butterfly on a tropical island. After receiving the first output, he continued to tweak his inputs. The prompts and resulting images are below:

An AI-generated image created using OpenAI's ChatGPT Images 2.0. The image was created with the following prompt: "Create an image of a chicken chasing a butterfly on a tropical island."

An AI-generated image created using OpenAI's ChatGPT Images 2.0. The second in a series of outputs, this image was created after the output of a previous image with the following prompt: "Make the image more photorealistic."

An AI-generated image created using OpenAI's ChatGPT Images 2.0. The third in a series of outputs, this image was created after the output of two previous images with the following prompt: "Remove the red flowers and replace them with blue. Make the butterfly a moth."

Hood described the process of creating the chicken images like "chasing your own tail round and round."

"If I need to generate an image and be able to edit and improve it, then I need better quality control and better continuity across the images," he said. "I don’t want to fix one thing to have it alter something else that needs fixing."

After the chicken, Hood gave the AI a more complex prompt. His prompts and the resulting images are below:

Two AI-generated images created using OpenAI's ChatGPT Images 2.0. The first image output, left, was created with the following prompt: "A realistic, high-fidelity editorial photo of a cluttered inventor’s desk in a dark, moody attic at midnight. In the exact center of the frame, there is a glowing, translucent glass cube. Inside the cube, a miniature forest grows with tiny bioluminescent trees. To the left of the cube, an open, weathered leather journal has the words 'THE GENESIS PROJECT' written clearly in a handwritten, elegant cursive script. Directly to the right of the cube, a half-eaten green apple rests on a stack of three thick, dusty blue books. The only light source is the intense, soft blue glow coming from the cube, which casts long, dramatic shadows and highlights the thick dust motes floating in the air. The background is slightly out of focus, showing blurred silhouettes of old wooden crates and a telescope. Capture this with the sharp detail of a 50mm lens, emphasizing the fine textures of the worn paper and the polished glass." The second output, right, was created with the additional guidance prompt: "Make it photoreal."

Concerns about deepfakes and revenge porn

Kathryn Coduto, media science professor at Boston University, hasn't used ChatGPT Images 2.0 but told USA TODAY she has concerns about how an improved image generator could create a higher risk for deepfakes and more specifically, revenge porn.

"I still think there's this high, high risk for this to be used in a really negative way, in a way that is particularly harmful to certain types of people or people in certain types of relationships," Coduto said. "With something that becomes this refined, you can truly have a custom-generated anything but that means you're having custom-generated anything without the consent of lots and lots of people."

Deepfakes, AI-generated media (photos, videos or audio) that resemble a person's likeness, have been on a rampage over the past year.

In early April, an Ohio man was believed to be the first person in the country convicted under a new federal law for using AI-generated, sexually explicit images of women to harass them. Prosecutors accused 37-year-old James Strahler of creating more than 700 non-consensual images and videos of adult women and minors, harassing the women with phone calls, messages and online postings. He will be sentenced at a later date, according to the U.S. Attorney's Office.

How much does ChatGPT Images 2.0 cost?

ChatGPT Images 2.0 can be used for free. Users without paid subscriptions are limited to two to three free images per day. OpenAI says the software is better with an OpenAI Plus ($20/month) or Pro ($100/month) subscription.

Greta Cross is a national trending reporter at USA TODAY. Story idea? Email her at [email protected].

Featured Weekly Ad