DALLE3 Gallery for 2023/2024: Share Your Creations

But here be the Princeps Peregrinorum, Director of the Roman CIA :laughing:

3 Likes

The OpenAI blog including DALL-E 3’s announcement and example has been decimated. Decimated might not be the right word, because “deci” as a root might imply there was even 1/10th the content remaining. Now the site is just corporate partner use cases back-dated into the blog and the DALL-E page just tells about safety considerations being taken.


So no more seeing if DALL-E will still do what it did from OpenAI’s own examples. Wayback machine also down probably due to another AI scraping campaign against it.


So I give you the best of 5 in the medium of polymer clay, with a high tendency to produce images with no content beyond the central box in others.

That has the signature appearance of having used the API parameter “natural”, which is actually symptomatic instead of natural: random objects dropped into the scene with no overall harmony of lighting or shadow, nor depiction of photographic depth or consistent sharpness. ChatGPT Plus users were getting an even worse version randomly for some of their generations.

3 Likes

The prompt:

"“Imagine a group of 8 explorers, women, men, and children, of varying ages and ethnicities, on a frontier, lush oil painting style, in the manner of a mid-19th century American landscape painter of the West. They have just reached a point in a mountain passage where they see a vast and beautiful land spread out before them, during the month of May.”

It had problems with number of explorers. And I had to go through some iterations until it got the multi-ethnic thing down. But the results were lovely. I ran this one through Photoshop to add an oil painting look.

I asked it about the inability to get the number of people correct. Fascinating responses.


"Generating images with an exact number of people can sometimes be challenging due to the complexity and subtlety involved in translating textual prompts into visual content. The AI uses patterns it learned during training to create images based on descriptions, but it doesn’t “count” in the human sense. Instead, it generates components of an image based on probabilities and associations from its training data.

When it comes to creating images with specific numbers of people, the AI might misinterpret the spatial and relational aspects of the prompt, leading to more or fewer figures than requested. Also, in complex scenes like group landscapes, the AI may struggle to maintain the exact number of figures while also trying to balance other elements of the composition, such as diversity and scenery. This can result in images that don’t perfectly adhere to the numerical aspect of the prompt."



"There are ongoing efforts to improve the accuracy and reliability of AI image generation, including better adherence to specific requests like the number of subjects in an image. Developers and researchers continually work on refining the models through updates and training with more diverse and detailed datasets. These updates aim to enhance the AI’s understanding of complex instructions and its ability to translate textual prompts into visually accurate representations more faithfully.

Further developments might include more sophisticated parsing of the text, better spatial understanding, and improved algorithms for image composition. These improvements could help the AI better manage specific requests like the exact number of people in a scene, making it more useful for precise tasks."


I asked it about its rendering problems when it generates a video for me.
Good stuff.

The AI model now called GPT-4 is simply a suggestible repeater of input, generating plausible completions.

(memory off; all tools disabled and empty custom instructions)

3 Likes

2 Likes



2 Likes

Welcome back. Hope you’ve been well.

blankz

3 Likes

2 Likes

2 Likes

1 Like




My friend who does podcasts with me made these in Chatgpt 4. Now known as Aria

1 Like





3 Likes

OP GOT OP

1 Like






1 Like

I call this “The Gift”

2 Likes




Another image of random happenstance.

4 Likes


we are waiting for u, meow ravers :slight_smile:

2 Likes





2 Likes