I Need help with what I am doing with generating consistent images through API using the Dalle-3 API specially in Pixar style. Pixar the word doesn’t work specially and then I tried alternatives such as the following:
In a 3D modern animation style, reminiscent of early 21st century design principles like Final Fantasy for children
It works consistent sometimes but it takes the old disney style sometimes please if someone can help in consistency for characters with their descriptions added in the prompts and images being generated un-supervised.
Thank you in advance Guys!
I have used the following:
I have tried my best to get the most similar and here are the results but I need help with having that generated in the API, also there is something interesting that I discovered, adding the following from the generated URL and the previous prompt had given me much consistency but I want to ask if someone else had this luck?
It has much more to do with inconsistent seed (read model interpretation) and randomized prompt rewrite than it does your actual input prompt. I’ve spent 100’s of hours prompt engineering Dalle, and I don’t think it’s possible to get a similar such style out every time (outside maybe some small niche of subject matter). No matter the prompt.
There seems to be some subject matter that the model is more likely to render in a given style whether it’s due to its training data or the words used to describe the subject I’m not sure.
If I’m getting inaccurate styles for a specific type of subject matter, I’ve had a little bit of luck slightly altering the words used to describe the subject, but I don’t think there’s a way around inconsistent styles simply through prompting right now.
Thank you so much for your response , and I would love if you could see my app that has a lot of generated stories through API with close to same character as shown. However, do you have any information on Azure OPENAI DALLE is it a fine- tuned version of DALLE?
