OpenAI is making a huge mistake by deprecating DALL-E-3

Thank you.

What model did you use?

This is my original prompt I used with DALL-E-3 in Microsoft Bing.

A cinematic, surreal, monochromatic image of a woman with metallic dark gray skin. Her bald head and shoulders are adorned with thorn-like spikes and twisting, vine-like structures. Flowing fabric swirls around her, shimmering with fiery stardust in hues of molten gold, ember red, and glowing orange. Faint burnished gold highlights glimmer along the vines, adding contrast to the muted gray tones. A textured, shadowy gray background with low-key lighting emphasizes sharp contours and intricate details. Convey a somber, intense atmosphere with an aura of mystery and photographic realism.

Someone replied to your post.

| windysoliloquy May 23 |

• | - |

Yeah you got some banger characters.

This is long. The little arrows like the one next to ‘start here’ are clickable so you can read the full contents of the message.

There are two ways to go about doing this. One gives you additional images of stuff you’ve already done quickly, the other helps the model understand your specific artist wants over time. Your sounding like you need the second option but both are demonstrated, in this session of mine:

Start here (click for more details)

And this link here will lead you to a session where I reverse engineered the prompt structure you need to recreate a given image. Once you get a given image recreated from the prompt itself rather than the image, you can then edit the prompt for something like location… You’ll be able to move the character to a new location and have it look consistent.

So don’t just learn the easy way… pay attention to how I get the model to cough your prompts up, and how I get it to refine them over a few more steps -

ChatGPT

ChatGPT - Reverse-engineering art style

Shared via ChatGPT

Prompt (click for more details)

Now when I get to the image above in the session I’ve given you… I can edit that image myself using chatgpt…

Say I want less vine type things growing on her, and deal with the need for a more intense light source… But I could send that as an image edit at this point because it’s that close to where your target style and character live.

Instead I went ahead and did one more round of reverse prompt engineering:

To show you that you can reverse engineer - too far. When you get close to your target style that you’re trying to import to the new model’s understanding… pay attention to when you’re close and then just do the standard image edits to walk you in to your exact space you’re trying to get to.

Because continuing the process as I’ve shown it to you too far starts working against you.

The easy way - (click for more details)

You would still have to edit the image to get a better contrast/lighting closer to your style for a direct hit.

But doing it the way I got this final image, is not going to help the model understand your artistic wants as well as reverse engineering the prompts.

The prompts are again, very large to start with, but as the model learns your particular artist styles and wants it will settle in and not require them to be so verbose down the road.