AI Look/Appearance Generator

[m.almohsin]

Hi VAM community,

I have recently discovered and dived into the Stable Diffusion Image Generator. This now being used to generate AI images using the computing power of your local computer (while some using it with cloud based computing such as Google Colab).

Depending on the trained model you’re using, you can generate some amazing human images of realistic or anime.

After testing it for myself, I thought this can be an amazing tool to generate looks/appearances inspired by images of actual people using this AI engine. Perhaps creating a specific model for VAM called (Image2Look)? I’m not a programmer, but I know someone in this community would love to take this challenge.

See below to get to know Stable Diffusion. Some of the most popular platforms that use this engine are:

• Stable Diffusion Automatic1111
• Stable Diffusion Web UI
• Stable Diffusion ComfyUi

Below some random examples of some of the extensions/workflows (text2image & image2image) that I have successfully used. I selected one of each platforms since they differ in their UIs.


(Example) Stable Diffusion Text2Image Generator:

(Example) Stable Diffusion Image2Image Generator:

(civitai.com) is a community hub that is being used for all type of amazing models to run using the Stable Diffusion engine. See example below:

Perfect World 完美世界 - v6 (Baked) | Stable Diffusion Checkpoint | Civitai

You can now run this model on SinkIn . The pursuit of perfect balance between realism and anime, a semi-realistic model aimed to achieve beautiful ...

civitai.com

I hope someone finds a way or ideas to integrate this to VAM for generating looks inspired by images of actual people.

Note: attached images are AI Generated using Stable Diffusion and are not mine, they were created by other individuals and were shared on civitai.com. These are very easy to generate. I created some amazing virts but are stored on my PC.

 

[hazmhox]

We have a hard time generating a proper 500 tris low poly house from a prompt at the moment. A fully compatible G9 character from a prompt is far from becoming a reality

 

[m.almohsin]

We have a hard time generating a proper 500 tris low poly house from a prompt at the moment. A fully compatible G9 character from a prompt is far from becoming a reality

It would be amazing to see how such a project develops. If only I have these skills.

 

[hazmhox]

"amazing", I don't know. That can be helpful.

But we already have dozens and dozens of looks published on a weekly basis. Having a "one button generation" feature would make unique creators like Juno, Damar, Ren, MonsterShinkai and all the great "manual" character creators feel like it's pointless to work hours on unique looks, when everyone upvote/react to one click generated characters.

 

[DJ]

If a "one button generation" feature can create looks that are anywhere near as good as the ones our quality creators spend hours making... well that's great for users! Though "vam look creator" will be added to the list of occupations replaced by AI.

But it seems to me you'd need an AI specifically trained on a set of morphs, so it could use those morphs to reproduce a reference image. That's a whole new AI project, not just an adaptation of an existing project, right?

 

[m.almohsin]

If a "one button generation" feature can create looks that are anywhere near as good as the ones our quality creators spend hours making... well that's great for users! Though "vam look creator" will be added to the list of occupations replaced by AI.

But it seems to me you'd need an AI specifically trained on a set of morphs, so it could use those morphs to reproduce a reference image. That's a whole new AI project, not just an adaptation of an existing project, right?

Great thoughts. That makes sense.

 

[AmazingDaemon]

I think there are already a lot of ways to do this but they require a few steps on a few different pieces of software. I’m not as familiar with Daz3D but there might be a way to from stable diffusion to Daz 3-D without having to go through facegen or blender.

That being said, these are relatively powerful tools with a lot of capabilities so whatever software package is eventually capable of this it’s not going to be just say press one button and you’re done sort of thing