It is possible with Stable Diffusion XL to obtain original and quality images in a relatively uncomplicated way. Here are 4 ways to do it…
In the world of text-to-image (AI image generator), three applications are vying for leadership, those that appeared in 2022 and sometimes even a little earlier:
- Midjourney is usually considered the best of the lot. Its handling is not easy. We have dedicated a tutorial to it on this page.
- Dall.e 2 is the text-to-image of OpenAI, creator of ChatGPT. It is much less advanced than Midjourney. However, it has the advantage of being accessible from the Bing search engine, as Bing Image Creator. As a result, a large number of users are discovering text-to-image via Dall.e 2. Again, we have published a tutorial for this application.
- Stable Diffusion is the work of the company Stability.ai and is accessible at the address dreamstudio.ai. However, since April 2023, a new version has appeared, Stable Diffusion XL. If we take care to use this exact version, we are then dealing with a tool of a quality close to Midjourney, with a slightly simpler approach.
Other applications have appeared since and some seem able to compete with the tenors of the field, in particular leonardo.ai or Adobe Firefly. However, access to these applications requires registering on the site of the corresponding publishers and waiting to receive an invitation, which can be very long.
So, Stable Diffusion XL appears to be a good solution if you want to obtain beautiful artistic achievements without having to become an expert in ‘prompt’.
Here we will cover 4 ways to get the most out of Stable Diffusion XL.
What is true for one is true for another
First, know one thing. If you know how to formulate a prompt on one of the text-to-image applications, then this know-how can be transposed elsewhere.
- Thus, in the Midjourney tutorial, we see how it is possible to enrich an image by indicating shadows and lights, color palettes…
- In the Dall.e 2 tutorial, we encourage you to indicate styles: art deco, engraving, vector…
All these tips are applicable to Stable Diffusion XL and for that matter also, to any text-to-image.
The essential point is that it is possible to obtain a quality image under Stable Diffusion XL, with a relatively simple prompt. Thus, since dreamstudio.ai, we simply typed a very simple request: “the most beautiful woman” (the application deals with prompts in English). Click under “Styles” and select one of those offered by Stable Diffusion. At the bottom of the screen on the left, make sure you have chosen: SDXL Beta Preview. Then click Dream. 4 images are then generated on the right side. It’s as simple as that, and the result should be of quality.
If you like one of the images, you can click on it to enlarge it and Generate variations (the 1st option). When you’re happy with an image, choose Download (the option on the right). Under Upscale, choose x2 to double the resolution of the image.
Another way to get a nice gallery picture is to go to the gallery Lexica.art dedicated to Stable Diffusion. Image creators regularly place their finest works there. And you can copy the prompts – even if it means adapting them to what you are looking for.
One of the attractions of Stable Diffusion is that this AI serves as an ‘engine’ for many applications: Night Café, Clipdrop, Replicate, Dreamlike.art… Quite often access is free for a limited number of images. If you want more latitude, you can subscribe to a subscription, which is usually inexpensive. Sometimes also, the application offers to earn free credits over the days. There are dozens of such apps. We like Night Café because of the pre-made templates it offers, but you might prefer Clipdrop which for its part, accepts prompts in French.
Under Night Café, be sure to select the SDXL BETA template. Specify even a simple prompt and choose one of many templates. Once an image appeals to you, click on Enhance to enlarge it. All that remains is to download it.
From a picture
Another option, you can from Stable Diffusion XL upload one of your photos. (Upload image). Then indicate a prompt – you can click on the Shuffle prompt button on the right – Stable Diffusion then offers you various prompts. It is important to influence the Image strength parameter – the higher the percentage, the closer the final image will be to the original. A percentage between 50 and 72% often seems adequate.
#Stable #Diffusion #properly