a few months back somebody posted great results to this forum using bing.com/create and the prompt "woman made out of slime". I changed it to "woman made out of paint" and went to town, see attached pics
Unfortunately, two weeks ago Bing.com switched to a new engine called "Dall-E 3", and now it's very difficult to create images with great paint coverage (images still look great, but the coverage is lacking). See the attached pics of the very same prompt.
Has anybody already figured out the prompt to create images where paint covers the whole face?
I think since the first word is "woman", it is very likely to generate facial characteristics that are not messy.
My suggestion is to try and move the word woman further down in the sentence. Here are some results I got from changing the order and words around a bit. The prompts are:
1. "head made out of paint and green goop, professional photography, female"
2 & 3. "head made out of black paint and green goop, professional photography, female, park background, leaning against a tree"
4. "head made out of black paint, body made of black goop, professional photography, female, park background, leaning against a tree"
Edit: My advice might be wrong. In stable diffusion, the order is important. I'm not sure about Dalle. When I changed the prompt to add a smile, it would often generate clean eyes and a clean mouth.
Mikee said: Fantastic suggestion, mFeelzGood - it works a treat!
I did previously experiment with order on bing creator / dall-e, and I can confirm that prompt order *does* matter.
Many thanks, that was really helpful!
Now I wonder if there is a service with the quality of Dall-E 3 and no censorship. Or is there a strategy to bypass the bing censor?
Yeah I wish Dall-e3 would become unrestricted. It is really good. I struggled to create something with stable diffusion, and dalle3 gets it in the first try. Dalle and midjourney are both private. Stable diffusion is open. That is why there are tons of websites that are popping up to allow you to generate with stable diffusion. It could be challenging to get good wam results with stable diffusion though. Hopefully the technology and models for stable diffusion improve.
The strategy to get around the censor that others mentioned in other threads is to use synonyms and words from different languages.