I need some guidance. I’m fairly confused. I have been using ChatGPT Plus for a little over a month. The main reason I signed up was because of its new image generating model. I had been using Ideogram. But i saw a bunch of YouTube videos that indicated ChatGPT now had a more advanced image generator (compared to Dalle-2) especially with regards to text within an image.
I tested ChatGPT a bit with a free account and I found that it was better than Ideogram, at least in my tests. So I signed up for a paid account. I have been using ChatGPT for non-image-related tasks but today I tried using it to generate some images and the results were poor. It had trouble with accurate text and it even had trouble re-generating an image when I asked it to regenerate some images with minor tweaks.
I’ve also seen some videos of people using an inpainting feature within ChatGPT but I don’t have access to it.
Anyway, I’ve been getting very frustrated with ChatGPT and today it kept referencing its image generation model as Dalle-3. I thought it was done using Dalle and that Dalle had been replaced by the newer, better image model.
So I don’t know what’s going on. I also went to Ideogram and did some tests and Ideogram definitely did a better with the images: Specifically, I wanted a photo of a room for a birthday party including a banner with specific words.
So now I’m thinking I need to go back to Ideogram, at least for image generation. But I’m still confused by ChatGPT. Has the new image model that does accurate text been rolled out to most users? Am I still using an older image model (Dalle within ChatGPT that is not great for text?
Same here — I’ve seen those YouTube videos too, especially ones that show inpainting and editing right inside the chat. But I don’t see that option in my interface either. Makes me wonder if they’re doing a slow rollout or if it’s only for enterprise users.
polepole, thanks so much for responding! I will look at some of those links. I have a Plus account and I’m using ChatGPT via the desktop app for macOS. ChatGPT keeps saying to me that I’m using Dalle-3, which is why I’'m so confused. I’m not using a custom GPT. I want to be using the latest image generation model. In additon, ChatGPT is doing a poor job regenerating an image with a tweak of the text in that image.
I need to go to bed so I’ll revisit this thread tomorrow, but I just clicked on an image, in the web browser, that ChatGPT generated for me a few hours ago and for that image I’m unable to use the selection tool. When I click on the image, I just see a bigger version of it, but there are no other options.
Sorry, one more reply: So, as I said, looks like the web browser version of ChatGPT is more full-featured than the desktop app. But I also tried using the selection tool and selected parts of an image and entered what I wanted changed. An ChatGPT generated a new image, but it slightly changed the details of the entire image, not just the selected part. So apparently the selection tool isn’t an actual in-painting tool. I was expecting to see a change just in the part I had selected.
ChatGPT uses '4o Image Generation`, not DALL-E, so it can create text better than DALL-E.
But, if there is too much objects and text, it can make mistakes.
Something else very frustrating, which makes ChatGPT worse than Ideogram and other more conventional AI image generators: I had been asking ChatGPT to generate images in a 16:9 ratio, that is 1080p. And I thought it was doing it, only to find out the images were a slighty different ratio and I hadn’t noticed it. So I asked to regenerate one of the images in a 16:9 ratio and again it gave me a different ratio (3:2)! It’s very frustrating. Is it even capable of 16:9?
I’ve been using ChatGPT intensely (for many hours) over the last two days. I had it generate a number of images for me, which took a lot of frustrating back and forth. I asked for images in a 16:9 aspect ratio, which it claimed it was doing. I trusted it as it’s easy to generate images in that ratio by other AI sites that generate images. But then I found out that the images it had generated for me were in a 3:2 aspect ratio! I now keep asking it to give me 16:9 and it keeps giving me 3:2! Before I signed up for ChatGPT, I had watched a number of YouTube videos extolling ChatGPT’s ability to generate accurate images from prompts, but none said that it has trouble with widescreen images.
Thanks for responding. DALLE doesn’t have the same accuracy as 4o. I’m completely surprised that none of the reviewers I saw on YouTube mentioned that ChatGPT’s new image generation model (4o) can’t do 16:9. To me, it’s a big deal.
