Amuse 2.0 - beta - AI image generation

[Trunks0]

Introducing Amuse 2.0 Beta with AMD XDNA™ Super Resolution: a fully local, AI experience

AI image generation is one of the most popular use-cases for generative AI and a quintessential AI PC experience is simply incomplete without it. Most of this local image generation happens using Stable Diffusion models created by Stability AI – which is a pioneer in this field. Multiple...

community.amd.com

System requirements are a little funky for this beta. Hopefully later versions will have wider HW support.

 

[DeathKnight]

Here's a few quick ones. Ran on balanced quality.

Generation is pretty fast on my 7900XTX. No XDNA super resolution though with just the graphics card and lack of compatible CPU.

 

[Trunks0]

Weird thing to not let the Super Res tech work on GPU. But it is a beta release, so maybe that will change.

 

[nutcrackr]

wow very cool, thanks for the heads up. It had to download 11gb model

"red miata driving through a busy street in pakistan" (for KAC)

 

[nutcrackr]

Getting lots of errors "GPU will not respond to the command" on my work laptop (RTX 3050)

I should try this on my home PC.

 

[DeathKnight]

A few more:

 

[Trunks0]

I generated some nice Nebulas and they responded quiet well to also being upscaled using "SwinIR BSRGAN x4"

I think I'm gonna rather enjoy using this to make backgrounds that work well for my 32:9 display.

 

[nutcrackr]

What model are you guys using? I tried to do a few people prompts and it the person had messed up face / hands / feet.

 

[Trunks0]

What model are you guys using? I tried to do a few people prompts and it the person had messed up face / hands / feet.

Suggest adding negative prompts. Like ugly; distorted; ugly eyes and etc. But it can often take a multiple generations to get a good result. Including a random name also helps for reasons I don't get when generating people lol.

Use ChatGPT to create prompts for Amuse. Make sure to ask for a prompt related to the model you are using (amuse says what the model in use is under the name). So like "Can you create a prompt for StableDiffusionXL?" And then ask if it can make for one for *insert your description of what you want*. This works well, because your using ChatGPT extremely good LLM to create a prompt designed for the far less impressive language model used by StableDiffusion models designed to run on our little systems.

I've liked the results with EpicRealism v5.

 

[RAY16]

Played around with it for a bit mostly using simple prompts I thought might be amusing. I think the novelty of that has worn of, though, and I doubt I'll be digging into it enough to generate more complex imagery. I might try generating a few images to use as wallpaper before I uninstall it, though.