Apr 04, 2023 12:30:00

A beta version of the image generation AI ``Stable Diffusion XL'', which is significantly stronger than the conventional Stable Diffusion, has been released as a test, so I tried using it

It has become clear that Stability AI, an AI research company that developed Stable Diffusion, is testing ` ` Stable Diffusion XL (#SDXL) '', which has greatly enhanced conventional image generation AI. I was able to actually use the beta version of 'Stable Diffusion XL' on Stability AI's web service '

DreamStudio ', so I tried to see how good the new image generation model was.

Stable Diffusion XL #SDXL is in testing with our partners! A model larger than the conventional SD of 900 million parameters, SDXL has 2.3 billion parameters and is improved by various methods ♪ https://t.co/tm7uzR7FNI
— Stability AI Japan Official (@StabilityAI_JP) April 1, 2023

The number of parameters included in the conventional Stable Diffusion was 900 million, but the number of parameters in 'Stable Diffusion XL' has reached 2.3 billion. The beta version of 'Stable Diffusion XL' is available in ' DreamStudio ', Stability AI's web service.

When accessing DreamStudio, at the time of writing the article, a popup was displayed announcing that early access to 'Stable Diffusion XL' was available as follows. Click 'Get Started'.

If you do not have a DreamStudio account, you will need to register a new account. Click 'Login' in the upper right.

You can create an account with a Google account, Discord account, email address, etc. This time, click 'Continue with Google'.

Select the account to use.

Click 'Accept'.

Account opening is now complete. DreamStudio gives you 25 credits at the time of initial registration, which allows you to generate about 125 images with default settings, so you can use this free credit to try 'Stable Diffusion XL'.

DreamStudio offers several Stable Diffusion models to choose from, including the beta Stable Diffusion XL. Click 'Model' on the upper left and first select the conventional version 'Stable Diffusion v2.1 (version 2.1)'.

This time, I will slightly modify the prompt found in '

unprompt.ai ', which allows you to search Stable Diffusion and Midjourney prompts from photos and illustrations, and generate an image with the default seed value and number of steps. After entering the prompt, click 'Dream'.

You will be asked to confirm

the terms of service , so check 'I agree to the Terms of Service' and click 'Accept'.

Then, 4 images were generated.

Looking at the four images side by side, it looks like this.

Next, change the model to 'SDXL Beta (Stable Diffusion XL Beta)' and try to generate an image with the same prompts and settings. Please note that the number of credits required is more than doubled compared to 'Stable Diffusion v2.1', so attention must be paid to the amount of credits consumed.