ASUS Agent Computer: Can Your Gaming PC run Local AI and LLMs?
Stable Diffusion 1.5 and XL
Stable Diffusion 1.5 and XL
If you’re aware of AI at all, you’ll be aware of it’s generative image qualities. Take a prompt, generate an image, job’s a good ‘un. Your Auntie probably uses it to make the whole family look like unicorns. World leaders use it to make their pools look clean. The titan of the user-friendly image generation world is Stable Diffusion. Obviously passed by other models in quality, but still the most famous with a wealth of models and LoRAs available. With the v1.5 test you get a 512 x 512 px image rendered in 100 steps, with four images per pass. If you prefer a higher resolution image and own a really meaty system you can utilise the XL model to get 1024 x 1024 px images.
Procyon gives us both a final score, akin to 3DMARK, but also times each generation. As you can see, despite only generating a single image instead of four, even if the total number of pixels generated at the same as 1.5, the XL model is complex enough that it takes 8 times as long per image. It’s why you can’t just look at raw “1 million pixels to generate” assumptions and guess the final result.


