OpenAI might have kicked off the text-to-image era craze with its DALL-E mannequin, however since these earlier glory days, the AI firm’s providing has been lapped by far more succesful picture fashions. Consequently, when OpenAI launched its newest and best GPT-4o picture era mannequin, I used to be skeptical. After testing it, I’ve modified my thoughts fully.
Getting began
When DALL-E first launched, it lived on its standalone web site; since then, it has moved to ChatGPT. The transfer got here with many advantages, together with having the ability to ask the AI chatbot for a picture you need in the identical interface the place you are already chatting about one thing else, thereby eliminating the necessity for fixed context switching.
With the discharge of GPT-4o picture era, OpenAI saved this handy format, switching the default picture generator from DALL-E to GPT-4o for paid subscribers. Consequently, it was tremendous straightforward to begin creating new pictures from my ChatGPT Plus account. All I needed to do was enter the immediate for what I needed to see, after which it will generate them. Customers may also entry it from the Sora interface.
Beware: You may nonetheless generate pictures equally in case you are a free person. Nevertheless, in the event you’re unimpressed, that is as a result of although at launch, the mannequin was introduced to be coming to all customers, together with free ones, OpenAI CEO Sam Altman introduced a day later that the rollout to the free tier would now be “delayed for awhile.”
The photographs
The second you might have been ready for — the photographs. After you insert a immediate, the AI outputs the era in below a minute. The method does take a bit longer than it used to, however the pictures are definitely worth the wait, delivering plenty of particulars, texture, realism, and even textual content accuracy. As a substitute of describing it, I’ll embody examples beneath so you may see for your self.
Immediate: Are you able to generate a sensible picture of a chameleon, up shut, shot as if it had been in Nationwide Geographic in 16:9 ratio?
Immediate: Are you able to generate a picture of a laptop computer open on a desk that claims, “This mannequin is so good that it could actually even get textual content and arms proper, that are normally main challenges for AI fashions,” with arms typing on a keyboard in 16:9 ratio?
Immediate: Are you able to generate a sensible photograph of a close-up of a girl in a crowd in Occasions Sq. trying on the digital camera and smiling, with the standard of 1 taken on a DSLR?
As seen above, the picture generator does an ideal job of adhering to the immediate and delivering high-quality, real looking pictures. Nevertheless, when testing an AI mannequin, one of many true efficiency metrics is the way it compares to opponents in the marketplace. To present you an excellent indicator of this, I made it generate the identical immediate I examined throughout the entire main AI picture mills, together with Midjourney, Google’s Imagen 3, Adobe Firefly, and extra.
I’m attaching GPT-4o’s rendition beneath. You may see the way it fares in opposition to the entire different AI picture mills on this article, together with DALL-E’s rendition, which clearly is way behind what the brand new mannequin can do.
Immediate: Are you able to generate a picture of a vibrant, real looking hummingbird perched on a tree?
Different notable options
Although the standard of the photographs is probably one of many mannequin’s greatest wins, there are different advantages as effectively. One of many greatest is that it lives within the chatbot’s interface, which makes it straightforward to tweak the generations with easy pure language prompts. Additionally, as a result of the chatbot has the context of what you simply requested it, it could actually take into account that in constructing the picture.
For instance, in case you are chatting with it about throwing a party, you might be able to say, “Are you able to now create an invitation that has the data above on it?” as a substitute of getting to retype. For instance, I began chatting with ChatGPT about throwing a housewarming, and when asking to make it create an invitation, I did not should repeat the data I beforehand mentioned.
You may also add reference pictures after which ask ChatGPT to create a distinct model or use them as components of a brand new one. For instance, you may enter it as a selfie and have it generated in anime model, as seen in Altman’s new X publish.
modified my pfp however perhaps somebody will make me a greater one
— Sam Altman (@sama) March 26, 2025
All of those customization options make it a very sturdy providing for creatives, who may also request that it’s rendered on a clear background or incorporate model model guides corresponding to hex codes or logos.
Talking of Altman, I used to be in a position to generate a picture of him sporting a celebration hat. I may achieve this as a result of the brand new mannequin has a lot looser safeguards, meant to permit customers to lean into their inventive freedom. The weblog publish saying the mannequin famous that it limits what will be created when actual persons are within the context, together with “significantly sturdy safeguards round nudity and graphic violence.”
I am unable to inform if there’s a sensible use case for this characteristic, however it’s a notable change I wanted to check out for myself. After I tried to create a picture of Mickey Mouse, it mentioned it could not resulting from copyright implications, so it appears not all public figures are honest sport.
Total
Total, the GPT-4o picture generator is a giant win over the DALL-E fashions and maybe among the many better of the various I’ve examined. Is it definitely worth the $20 per thirty days? In case you are simply eager about high-quality picture era, there are nonetheless free variations you may discover which might be actually succesful, corresponding to Adobe Firefly or Google’s Imagen 3.
Having mentioned this, in case you are a frequent ChatGPT person, the improve to ChatGPT Plus will get considerably extra engaging. With this improve, you’ll have entry to all of OpenAI’s newest and best chatbot options, in addition to high-quality picture and video era, all for $20 a month, which isn’t a nasty deal, particularly contemplating different choices in the marketplace. For instance, Midjourney’s subscription begins at $10 per thirty days and solely gives picture era.
Need extra tales about AI? Join Innovation, our weekly publication.
