OpenAI’s New Image Generation – Game-Changing Quality

It’s crazy to think that around two months ago I was seriously considering canceling my OpenAI subscription. I was planning to keep the API open but cancel the front-end because I was so rarely using it. I was (and still am) using Claude to do all my writing, and I was (and still am) using Cursor for all my programming. So really, I was thinking, “Well, even though it’s $20 a month, I’m not really using it.”

And then they came out with Deep Research, and despite the high price, it really hooked me. So I thought, “Okay, well I’ll keep it going a bit longer.”

But now, within the past day, they’ve added to the old GPT-4o model the capability to produce amazing quality images. It’s a little bit slow in my usage today — it seems to be taking maybe around 45 seconds on average to make images — but the quality is insane.

For my first test, I gave it a copy of my business logo from Argelius Labs and asked it to make it 3D, and it did it perfectly. It did amazing. If I had paid someone to do that, I would have been very happy with the results.

Then I gave it the logos from two sites that are used in labs for SANS OSINT courses — RateMyPolarBear.com and RateMyPandaBear.com — and I said, “Hey, here’s the two mascots, make an image with them in a back-alley fight.” The images that it made were incredible. I think my favorite part was in one of the images where bamboo was laying on the ground, slowly accumulating over the course of multiple images.

Right now it has some limitations. It says it doesn’t want to do things of people, even if it’s you, but it’ll do it in a similar style. So I gave it a picture of me and asked it to turn me into a Lego person, and it still did a good job. I think the baldness is a little harsh, but that’s my problem — I’ve got to live with that.

The results here are really kind of insane. If you haven’t played with it yet, you should. I’m not going to lie — this is not a good thing for graphic artists because AI image generation has seriously been lacking until now. The text was horrible, there were some other problems. It was still really impressive at times, but far from perfect.

Now, seeing what OpenAI has come out with, I’m really blown away. Like I said, the things that I’ve done today playing around with it — if I had paid someone and that’s what they sent me, I would have been very, very happy with those results. So to get them included with the plan is kind of ridiculous.

If you have a chance to play with it and have any thoughts, let me know. I’d love to hear about your experience!