I’m reaching out to the community for some insights into an issue I’ve been facing with DALL-E 3. As a regular and subscribed user, I’ve noticed a significant drop in the quality of the images generated by the tool, especially after extensive use or when making multiple requests of a similar nature.
In the past few weeks, this problem has become more pronounced. Initially, DALL-E 3 produced stunning and high-quality images, but recently, the outputs have often been subpar, sometimes even reaching what I would describe as ‘catastrophic’ quality levels. This happens particularly after I’ve made many requests in a short period or when I try to generate variations of the same theme.
Has anyone else experienced similar issues with DALL-E 3? I’m trying to understand if this is a common problem or something specific to my usage. If you’ve encountered this and found a solution or workaround, your advice would be greatly appreciated!Thanks in advance for your help!
Best regards
Here are a terrible result after i asked for the same prompt than before (three animals antropomorphic talking out loud after a dinner, cartoon oil painting). Two first requests of the day. The first one went fine.
Yes the same has happened to me. When I first start using DALL-E it is great. The quality declines with each regeneration for me as well. Once I have been using it for about an hour EVERY image is terrible quality. I agree with “catastrophic”. I had a image so bad I thought to myself it was something I would see in my nightmares. DALL-E even asked me to take a break my generating. I was going to take a break anyway cause the terrible quality was frustrating lol
I’ve noticed this too. I have hundreds of really high quality images, however in the last three months the quality of all images produced has been so incredibly low I stopped using it. It’s terrible. Even when you ask it to increase the resolution, it just gets worse. Not happy. I pay for this. I expect a quality product.
I pay for ChatGPT and it includes a version of Dall-E. I stopped my subscription to Midjourney thinking what little I need could be done but OpenAi. As of today, it’s only good if you want something that looks like a cartoon. And forget about asking it to create art in the style of a specific artist. It won’t do that at all.
I have been experiencing this as well. It seems over the past week there has been a drastic decline in image quality from DALL-E. It doesn’t seem to matter how detailed the prompt is. It also doesn’t seem to matter if I am generating multiple images in a row, or generating a set of images off of a fresh prompt. The images are completely unusable. I would love to know if the developers are aware of this.
I’ve noticed the decline also, especially when it comes to generating faces. A couple of months ago it used to be really good, and now faces are distorted and deformed in some cases. The quality of the initial output seems to have declined greatly!
Couldn’t agree more. The ironic part is I can go to Microsoft Designer which also uses DALL-E 3, put in the same exact prompt, and the image quality is excellent. It is as good as the images that OpenAI’s DALL-E used to produce just a few months ago. The sad part is that apparently OpenAI doesn’t care about the quality of the images they provide their paying customers. If the results of my text prompts weren’t so good -other than the image- I’d end my subscription to OpenAI in a heartbeat. I will use OpenAI for text prompt responses and will look to another image generator (Microsoft?) for quality images.
OpenAI DALL-E
Grainy, lack of sharpness and washed out colors
Someone close my threat of Dalle 3 images lower resolution, with no response from the Dalle 3 team, solution or answers that’s frustrating so here is my image
I am experiencing the same issues, it’s very frustrating. The quality for the first couple hours was excellent and the consistency as well, but after 30 up to 40 minutes into the conversation, the quality started decreasing drastically my 2 years old toddler can scramble around better images
First of all, this is only a theory I have, as I have not been able to validate this. I think the problem might be with how underlying models are trained. I think DALL-E 3 on chat GPT is always using the generated data as new training info, so in time this will lead to a model basically being trained on images which are AI generated. Of course this would lead to a decrease in quality as we continue to use the image generation tool. I believe Microsoft uses a model that is not continually retrained, so even if it uses the same underlying tech, the model is probably not updated so frequently so quality does not degrade as fast (though I’ve noticed it is starting to degrade also, especially with hands and faces, is it gives errors akin to Stable Diffusion in their early models).
Truth is, as mostly I pay GPT Plus for the image generation, if the quality is going to be worse than free tools, I will definitely consider stopping paying and switching to other tools.
that’s frustrating so here is my image
