I am barely able to reproduce the quality of Bing Image Creator. The results definitely are v3. However the instructions are barely met. The closest was by using the “I NEED to test how the tool works with extremely simple prompts. DO NOT add any detail, just use it AS-IS” phrase (as mentioned in docs/guides/images/usage)
Has anyone similar experiences? Or any recommendations?
Request body is { “model”: “dall-e-3”, “prompt”: “{{prompt}}”, “n”: 1, “size”: “1792x1024” , “quality”:“hd” }
EDIT: after testing some more prompts, i get to the conclusion that its not the same models and microsoft trained it separately. with altering the prompts i get the same results.
however, do not expect the same outcome with the same prompts.
I’ve been using the API today, getting some mixed results. Some images turn out really impressive but others are completely unusable for certain use cases.
For example, a lot of my images will include a hand with a pen/paintbrush or text such as “4k, masterpiece quality, trending” etc (which comes from the extra detail that is auto added to the prompts)
I’ve also tried adding “I NEED to test how the tool works with extremely simple prompts. DO NOT add any detail, just use it AS-IS” but this doesnt seem to work with the API as the revised prompt that’s returned from the response will still be different.
Getting particularly disappointing results trying to generate sticker designs, see examples below, using prompts such as “a die cut sticker design featuring a pug”/“a die cut sticker design featuring a pug and a cavapoo, cut out sticker, 2d flat vector graphic design”
