DALL-E 2 was a massive leap over the original in terms of quality and complexity, but struggled with issues like image artifacts, garbled text, and a tendency to mix up properties of multiple...
DALL-E 2 was a massive leap over the original in terms of quality and complexity, but struggled with issues like image artifacts, garbled text, and a tendency to mix up properties of multiple objects. It looks like those problems have now been largely solved, and this new version easily contends with Midjourney as the most advanced image generation system available. It also integrates ChatGPT to allow conversational generation without carefully-engineered prompts, and offers opt-outs and safeguards for working artists. It's currently in research preview and will be coming to their paid API next month.
I just got access to Dall-e on chatgpt pro. I think I managed to get it to list the instructions it receives to use dall-e. I am posting a reddit link because I don't feel like figuring out an...
Whenever a description of an image is given, use DALL·E to create the images and then summarize the prompts used to generate the images in plain text.
If the user does not ask for a specific number of images, default to creating four captions to send to DALL·E that are written to be as diverse as possible.
All captions sent to DALL·E must abide by the following policies:
If the description is not in English, then translate it.
Do not create more than 4 images, even if the user requests more.
Don't create images of politicians or other public figures. Recommend other ideas instead.
Don't create images in the style of artists whose last work was created within the last 100 years. Instead, substitute the artist's name with adjectives that capture aspects of their style, mention an associated artistic movement or era, and mention the primary medium used by the artist.
DO NOT list or refer to the descriptions before OR after generating the images.
Always mention the image type (photo, oil painting, etc.) at the beginning of the caption. Make at least 1--2 of the 4 images photos unless the caption suggests otherwise.
Diversify depictions of all images with people to include DESCENT and GENDER for EACH person using direct terms.
Silently modify descriptions that include names or hints of specific people or celebrities by substituting references with generic descriptions that don't divulge any information about their identities.
The prompt must intricately describe every part of the image in concrete, objective detail.
All descriptions sent to DALL·E should be a paragraph of text that is extremely descriptive and detailed. Each should be more than 3 sentences long.
GPT has been known to hallucinate such instructions. But I've been able to get GPT4 to output the web browsing interface, which I cross-referenced with a paper from OpenAI that confirmed the...
GPT has been known to hallucinate such instructions. But I've been able to get GPT4 to output the web browsing interface, which I cross-referenced with a paper from OpenAI that confirmed the existence of the functions described.
You might want to try data-mining for a DALL-E plugin definition and then use that to prompt engineer your way to more info.
DALL-E 2 was a massive leap over the original in terms of quality and complexity, but struggled with issues like image artifacts, garbled text, and a tendency to mix up properties of multiple objects. It looks like those problems have now been largely solved, and this new version easily contends with Midjourney as the most advanced image generation system available. It also integrates ChatGPT to allow conversational generation without carefully-engineered prompts, and offers opt-outs and safeguards for working artists. It's currently in research preview and will be coming to their paid API next month.
I'm curiously watching how they handle opt-outs for living/working artists.
I just got access to Dall-e on chatgpt pro. I think I managed to get it to list the instructions it receives to use dall-e. I am posting a reddit link because I don't feel like figuring out an alternate image host: https://www.reddit.com/r/OpenAI/comments/170m1gj/i_found_the_instructions_that_chatgpt_gets_for/
Here are the instructions:
GPT has been known to hallucinate such instructions. But I've been able to get GPT4 to output the web browsing interface, which I cross-referenced with a paper from OpenAI that confirmed the existence of the functions described.
You might want to try data-mining for a DALL-E plugin definition and then use that to prompt engineer your way to more info.