It is not perfect, but it is better than before.
4 Likes
Yes, the skin texture is still not natural, but it is better then some months ago.
But in the same time, toxic data came in the training with more photo pictures.
3 Likes
A post was merged into an existing topic: Official 2025 Dall-E Mega Gallery 
Will dalle-3 become free and available since itās outdated and useless now?
Itās free as of 2023. Bing Image Creator
Yeah, but is that not manual? I mean for api calls its still 3cents.
I donāt know how Bing works. But thereās a daily limit of 100+ queries there. Thatās a total of 500+ images per day. And the censorship is not as strong as DALL-E GPT. But Bing has a kindergarten mode too.
1 Like
This is super helpful. Do you know if the new model gpt-image-1 fixed the issue of ānonsensical text insertionā?
1 Like
As much i can say, yes. The new image generator is a different System then the DallE3.
- No Nonsense text (as much i know, i have not seen any. if text shows up, it is because the system is confused if it has to insert text.)
- Orientation is unstable but now works for portrait format.
- Not to light anymore, no nonsense lights anymore in the scene.
But now it comes with new issues:
- Prompt goos, a prompt is influenced by the preview ones.
- Detail effect causes distortions.
- To dark, often images are to dark, specially if darkness is in the prompt.
- Less imaginative. DallE3 was more inventive.
And the dysfunctional security system still goes on every bodies nerves.
etcā¦
You can find more here:
1 Like
Thatās good to know! And from my reading, the new model gpt-image-1 takes longer to generate images. (Which is less preferred for the app I am building)
1 Like
Yes, at least if you use ChatGPT. I am not a API user, so i can not help with experiences from this part. But here in this thread, you will find tips for prompting and help if possible, for some of the issues.
But this thread is more focused on the generator it self. If you are focused on API development for the Image Generator, you can open a thread for it, if you not find any already talking about what interests you.
The thread for DallE-3 is still relevant for API users, because DallE 2 and 3 is still accessible. Just not for ChatGPT users (like me).
Here one to start withā¦
https://help.openai.com/en/articles/11128753-gpt-image-api
2 Likes
I havnt been able to produce any good quality images for almost 2 weeks! I was able to create really god images but now its all dull and boring and I keep getting error messages!!
1 Like
I felt the same. They are not made for the same consumers in my opinion. The new model is more for ADS and commercial use. DallE was more for artists. It is possible to get something out with more prompt text, but not like with DallE. The new generator is more photo realistic, and DallE more fantasy full. I liked DallE much more. We can just hope that a offline model once is trained correctly and hardware becomes affordable. And that maybe the new model will be trained better in the futureā¦
1 Like
So is dalle not available anymore? Iāve tried previous prompts in dalle and chat and non of them work anymore and I get a very basic image. Itās the main reason I continued to pay for the monthly pro lab.
1 Like
Yes, Dall-E is still accessible from the official Dall-E GPT under the āExplore GPTsā tab of the chat interface as well as via the API.
3 Likes
Thanks for the tip, I didnāt know that!
It is not accessible anymore by a MyGPT.
Iāve always used my own GPT, and DALLĀ·E is no longer accessible from the MyGPT options, but the official DALLĀ·E link actually still works (although I wasnāt able to find it through the search anymore).
3 Likes
One thing I noticed on the DALLE Custom GPT is that it outputs 2 images by defaultā¦
2 Likes
We tested the following prompt using DALLĀ·E 3 inside ChatGPT (no seed):
A glowing lotus flower emerges in a dreamlike, inverted-perspective cosmic field. Its petals radiate golden light, while the dark-violet void behind it adds depth. Delicate golden swirls form a fractal vortex spiraling toward the observer. A translucent feminine silhouette drifts in the distance. Light and shadow dance in a harmonic flow, with subtle depth shifts as if the scene breathes. Use atmospheric volumetric lighting, soft contrast, and golden highlights on cool cosmic tones. Render in surreal 3D painterly style.
We received two very different images, despite using the same prompt:
Image A: soft, flat, light ethereal
Image B: rich depth, strong inverted perspective
Question:
How do you stabilize DALLĀ·E outputs across generations?
Prompting tips for visual consistency?
Can a āfixed compositionā be enforced?
Any method to lock or reuse seeds within ChatGPT?
Thanks in advance for your feedback!
Nicholas
1 Like
Pretty sure thatās gpt-image-1 not DALLE3.
2 Likes
Echoing @PaulBellow, there is no DALL-E 3 inside ChatGPT.
Your image quality looks like '4o Image Generation".
DALL-E is a saperate custom GPT.
3 Likes