Description
Image imported from Derpibooru.
Original: https://derpibooru.org/images/3213318
Metadata at the time of import:
Original: https://derpibooru.org/images/3213318
Metadata at the time of import:
| Field | Data |
|---|---|
| Favorites | 90 |
| Upvotes | 129 |
| Downvotes | 3 |
| Score | 126 |
| Comments | 10 |
| Uploader | Anonymous |
Original description below this line
Creating using Dalle 3, available for free on Bing’s image generator here.
Prompt (approx.):
My Little Pony, Rarity exploring a dazzling stone cave with giant crystal shards of dozens of colors, her way lit only by the light blue glow of her horn, sparse faint sun rays from above, white unicorn, purple hair, blue eyes, professional impressionist digital art style
It is indeed possible that the wrong cutie mark was deliberately generated because of copyright issues.
I think so too, just because I have ChatGPT and seen how it writes your prompts for you after you tell it what you want, I also think there’s more settings and stuff not even accessible to the user, I mean of course, like steps and aspect-ratio and stuff like that but the prompts are the real important thing here, though I’m just guessing here.
I believe Bing AI pre-processes your prompts with ChatGPT or something similar before sending them to DALL-E 3.
The way OpenAI and Stable Diffusion work is very different I believe, Stable Diffusion seems to be open completely source and also doesn’t have any real copyright issues/concerns, as of Bing it has to worry about its users infringing copyright or generating things that shouldn’t be generated.
Granted I’ve never used Stable Diffusion due to hardware limitations although it does open the floodgates to other users as well or allows others to get their foot in the door.
I noticed that I didn’t look at the characters before, and there were really no “cutie mark”.
The (apparently approximate) prompt is in the description and it doesn’t describe the cutie mark. But I wonder how much would describing it help…
I see. I thought the prompt had clearly listed the cutie mark, and the resulting picture was still the wrong cutie mark
I think the AI doesn’t recall MLP characters well. You can see that the prompt explicitly lists Rarity’s mane, coat, and eye color. It also specifies that she’s a unicorn. Apparently her name was not enough. Cutie mark was not specified, so AI just slapped on whatever.