AI Image Functionality & Templates Discussion

Community Development
1

For shits and giggles I hooked up OpenAI’s Dalle2 image generation and edit APIs to Mimir. This is available via a slash command:

Command

image

Prompts

A prompt is a simple description about the type of image you are requesting. Some to try:

Templates

In addition to this general, prompt, you can choose to add a template. The command has several preset templates ranging from Kim Jong Un to Jerome Powell, to me and “JB”. Templates allow you to “build a picture” around an image. The presets currently are just faces with the exception of CCIV’s emoji which is an angry anime girl :kekw:

image
image512×512 91.7 KB

The image above was constructed by selecting the CCIV template and using the prompt anime girl surrounded by bags of trash:

image

Template…uh… templates

Below are two template files. You can create your own templates by cutting out transparent space around a person or character and positioning them in a sensical location in the whitespace (imagine where you want their head to be in the pictures you generate because DALLE doesn’t move it around and just draws around it.)

The size of the total image should be 512x512. I would suggest loading one of the files below into your image editor and just erasing the character there and adding your own. After you’re done, just post the image here and I’ll get it added.

Images of people in the community regardless of whether or not they’ve posted the photos themselves previously require their permission to be added.

cciv2
cciv2512×512 102 KB

kimjong2
kimjong2512×512 30.2 KB

Now back to doing meaningful dev work :pepepray:

2

One quick note while I’m thinking of it, photos looks straight forward actually work really poorly for whatever reason so if you’re picking pics of people, choose one where they’re looking slightly off to the side.

3

This received one small upgrade this evening. There is now an additional parameter called “Prompt Helper”. Since we’re all lazy sacks of shit and struggle writing detailed descriptions of the images we want, this feature pushes your shitty, worthless idea through GPT and pumps out something detailed and actually, well, good.

For example my shitty stupid prompt of:

brunette pilot in cockpit

Gets unshitty’d and turned into:

A stunning brunette pilot sits in the cockpit of a futuristic spacecraft. She surveys the endless stars of the cosmos as she navigates towards an unknown destination. A plethora of buttons, switches and levers fill the console before her. Outside the craft, the vacuum of space radiates its boundless energy, creating an otherworldly aura around her. She takes a deep breath and with a determined glint in her eye, starts to pilot her ship deeper into the unknown.

Which much to nobody’s surprise pumps out an actual image and not some jumbled mess of bullshit like we’ve been requesting for days now:

Image

In order to use this feature simply set prompt_helper to Enabled. It works for all models and it feeds your expanded prompt back to you just in case you’re wondering what amazing shit it asked for on your behalf.

Happy fucking around :pepecelebrate:

3 Likes