Use this link ChatGPT - DALL·E
3 Likes
Yes, mine too! The quality is dreadful, exactly like a child has created them!
1 Like
i have wrote why i dont like dalle, i know i can use it…
Me too. In addition most of innocent images I request are banned as “violation” without any possible logic
2 Likes
Looks like something changed in the last few days… quality is not the same as launch. I asked Chatgpt and got this reply.
“Yes—while there hasn’t been an official announcement about a major change, there have been noticeable shifts in image generation behavior over the past few days, especially regarding:…”
That said, you’re absolutely not alone. A lot of people noticed this shift, especially those working with collage, comics, or anything that needs a faithful image-to-image style swap. I can flag your feedback so it gets heard by the team—it genuinely helps push for better options and maybe even a rollback or alternate mode.
2 Likes
Me too? Same prompts. First one etheric now a stupid ugly impressionist image or sad faced girl in a coffee stained dress. All images looking like they belong in a dark century cathedral. I use to get stunning images. Now they are flat, no life, just dreadful.
1 Like
hi, we are March 31 now is there any update info about this image quality loss?
no updates that I can see, but personally, it sounds like they dialled it down quite a bit. I was actually pleasantly surprised with the update, I was getting a good output for what I actually asked, but since the Studio Ghibli issues became a problem, I feel like they did something that completely messed up the model and now all the images lost quality, I’m getting fuzzy, dark weird looking images which is a huge let down compared to what I was getting 3 or 4 days ago
Been using the new model for some time now… mixed feelings as far as quality is concerned, not fit for purpose for me anymore… I am generating historical images that need a lot of tweaking in order to get rid of the obvious hollywoood crap that the model was trained on. In order to have one usable image I always needed like 10 to 20 images. maybe another 3-4 edits in the image I selected to use. This used to require maybe half an hour. Now I need an hour to create 10 images, the generation time is abhorrent, although it got better in the last two days and the limits now are insane. With a Plus subscription I now need to wait 6-10 minutes to generate a new one. The photorealistic detail, textiles, faces as well as certain parameters like number of items, colour etc have improved. However, prompts are now almost randomly generated, I now ask the system to have me confirm the prompt before generation always, even if I just want something specific to change (like have a person wear a blue tunic instead of a red one). It seems that the system just generates some kind of summarised prompt adding its own little details that screw things completely (for example chainging photorealistic to oil painting style…). I also observed that generation of crowds (I have to create armies for example) has decreased in quality a lot. Now, stances are not as natural or diverse, I have 10 men for example with the exact same face, garments or even stance. I also have to give much more specific prompts regarding lighting, colours etc. Restrictions seem to now be even more insane, I could not have the system draw a frightened child looking at soldiers in the distance… In effect, I am now struggling 20 hours already and I have not even done the work I used to need 4 hours for. I really hope all this is passing and that things will return to normal but I doubt it… I only hope that they would allow us to use the former model with the former limitations and before you say we can, no, we cannot. What one accesses with the DallE app has nothing to do with what was integrated into ChatGPT. In all, I am very dismayed. No matter what people say in comparisons, not Leonardo nor Midjourney were ever as good for historical content generation as DallE integrated in ChatGPT.
“I want. I want an image to look realistic yet painted.”
However, perhaps you’re referring to the art style rather than the description? I analyzed the image to identify its specific artistic style, and it is best described as: Hyperrealistic Digital Painting (with a Painterly Twist)
Digital Painting Techniques:
- The clean lines and smooth gradients suggest it was created using digital tools (like Photoshop, Procreate, or AI-assisted digital painting software).
If that is the stylistic but not the description then it can only be done with DALL-E,
however, based on your Description
I tried your image on Sora, and it seems to work perfectly with the new image generator released by OpenAI. Here are the images I created based on your description of what you’re trying to achieve:
Original
Gen4
I agree that DALL·E should remain an option on Sora—paying users deserve to have choices. Perhaps OpenAI will eventually clarify or expand user options. While DALL·E’s unique digital art style isn’t exactly my taste—or should I say, my cup of tea—it still holds value, even if it can’t always be used as a direct reference. OpenAI should include a simple toggle or option on Sora to let people choose whether or not they want to use DALL·E. Besides, other competitors will eventually catch up to this incredible technological breakthrough, and they might offer similar or even better tools, making the space more competitive. That said, I have to admit, I’m genuinely impressed with the new image generator—it’s surprisingly addictive.
UPDATE: DALLE-3 can now be found on Custom Chatgpts. Just add the plugin.
1 Like
All the hype is annoying, chatgpt sucks now
did u find any alternative..i am also experiencing the same issue
1 Like