Happy New Year guys, I was wondering if any of you are creative enough to help me come up with prompts for these photos. I use Bing AI and have been getting hit with the confused dog for all of them

More references guys please help

Good luck with your Camera POV project! It really seems interesting!

Cheers,
Team pregR

(January 2, 2024, 11:36 am)TeampregR Good luck with your Camera POV project! It really seems interesting!

Cheers,
Team pregR

Thanks, I just want to give you guys some content is all :/

Best advice I would give is look at Shutterstock or similar stock photo sites, their descriptions are pretty de-sexualized enough that generations are unlikely (but not a given):to trigger "the dog".

Bing AI assistant/chat component (the chatGPT part rather than image generation) can actually provide descriptions of source images that you upload, which can be used for prompting the Dalle component.

Course even using a prompt that chatGPT generated doesn't mean the dog won't pop-up, as it's totally dictated by what image is generated.

Blame Microsoft for that, it's them that apply it rather than Dalle3/chatGPT. Sometimes even the most wholesome of content will get flagged if the generated image includes something the filter thinks is weird.

Always remember, the dog doesn't mean the prompt is bad per se (if the prompt has a banned concept, you'll see a different error than the dog) only that the image generated contains something the filter doesn't like. Precisely why the more it looks like a stock image, the better likelihood it'll be viewed as wholesome.

(January 8, 2024, 9:48 pm)mimmiberg Best advice I would give is look at Shutterstock or similar stock photo sites, their descriptions are pretty de-sexualized enough that generations are unlikely (but not a given):to trigger "the dog".

Bing AI assistant/chat component (the chatGPT part rather than image generation) can actually provide descriptions of source images that you upload, which can be used for prompting the Dalle component.

Course even using a prompt that chatGPT generated doesn't mean the dog won't pop-up, as it's totally dictated by what image is generated.

Blame Microsoft for that, it's them that apply it rather than Dalle3/chatGPT. Sometimes even the most wholesome of content will get flagged if the generated image includes something the filter thinks is weird.

Always remember, the dog doesn't mean the prompt is bad per se (if the prompt has a banned concept, you'll see a different error than the dog) only that the image generated contains something the filter doesn't like. Precisely why the more it looks like a stock image, the better likelihood it'll be viewed as wholesome.

You my good sir are any evil genius! Thank you although I have yet to successfully generate the couple photographs…

Found something that might be easier than Bing Chat for getting a prompt, or at least descriptions:
hxxps://www.astica.org/vision/describe/

Specifically the GPT-S description, course it's too terse to use the entirely as a prompt, but you can pick out some sentences.

Here's an example that worked 4 of 5 times (only number of times I tried) with the same prompt:
Src Image:
   

Prompt:

Quote:A pregnant woman standing in front of a window, holding a glass of green juice. She is wearing a garment that shows off her baby bump and has her hand resting on her stomach. The woman appears to be in her early twenties and has long brown hair. Her face is turned slightly to the side as she looks down at her phone, which she holds in one hand.

---

               

Not sure why it thought she had a phone?

Some others from the same prompt but tweaked slightly.

           

Then used the word "muscles" thinking it might "wholesomely" include some skin.... oh hello dog.

Back to stable diffusion I go, though it's probably worth noting you can use Dalle 3 on GPT 4 without Microsoft's bastardized filtering (it still has some of it's own filtering), of course for a cost.

Tempted to give it a try and see how true that really is.

Turns out Dalle 3 on GPT4 is even worse, even when I told it to use a prompt that completely fits it policies and guidelines.

Tried something simple and this was the result:

   

Like come on, even changing it from belly -> abdomen resulted in nada.

@Em8arazada this might cheer you up from the depression that is Bing Image Creator

           

Some animated action