Hello everyone~! Are you enjoying your day today?
Today, I want to share some exciting news with you. It’s all about SDXL Turbo! Let’s explore together what SDXL Turbo is all about, shall we?
SDXL
SDXL is an abbreviation for Stable Diffusion XL, representing the latest advancement in the text-to-image model series, Stable Diffusion, introduced by Stability AI. This robust algorithm has the capability to create high-resolution, realistic images from various text inputs, including descriptions, captions, keywords, and even simple sketches.
1. What is SDXL Turbo?
SDXL Turbo is an innovative feature that enables real-time image generation from text, eliminating any waiting or loading times.
This is the key aspect of SDXL Turbo, appealing profoundly to all AI artists. It utilizes a specialized version of SDXL that’s fine-tuned for swift performance and speed, all while maintaining high quality and diversity. You might remember my post about LCM that can be used with LoRA; SDXL Turbo can also be downloaded and used in a similar fashion.
–> Go to Post (What is LCM LoRA and How to Use)
So, let’s skip the long description and get straight to how you can use it, which I’m sure will be more interesting for you!
2. How to use SDXL Turbo
First, to use SDXL Turbo, please download the model through the link below!
After downloading the model, place it in the Stable Diffusion>Models>Stable-diffusion folder and then launch WebUI Stable Diffusion.
** Oh! If you don’t have Stable Diffusion installed, please refer to the following post for installation instructions!
–> Go to Post (How to install Stable Diffusion)
Once you’ve completed all the steps, go to the Txt2Img tab and select the Turbo model you’ve just downloaded as the Basemodel. The Base Model can be chosen from a dropdown menu in the top left corner, as mentioned earlier.
If you’ve placed the downloaded model file in the Models>Stable Diffusion folder, it will be visible.
Next, we’ll write a simple prompt to explore how it can be used. For those who are new to this, please follow the settings I’ve listed below.
Settings
- Base Model : sd_xl_turbo_1.0_fp16.safetensors
- Positive Prompt : A chubby cat
- Sampling Method : Euler a
- Sampling Steps : 1
- Size : 512X512
- CFG : 1
Now, let’s click the Generate Image button on the right and see what kind of outcome we get!
Firstly, we encounter an astonishing fact here. Can you see the incredible speed of the rendering?
Unsurprisingly, given the simplicity of the prompt, the quality isn’t great… haha.
Next, we’re going to incrementally add to what we wrote in the Positive Prompt. Let’s observe how the Generating Image on the right evolves with each new word added.
The setting we need to interact with is ‘Generate Forever.’
Given SDXL Turbo’s incredibly fast rendering times, which allow for real-time adjustments to prompt values, this feature is a major benefit.
We enable the ‘Generate Forever’ setting by right-clicking the Generate button on the right. Be aware that the gauge bar will appear as not fully filled, as illustrated in the picture below.
As demonstrated in the photo above, once the Generate Forever feature is activated, new images will continuously be generated based on the specified prompts and settings. Because we are using SDXL Turbo as our Base Model, images will be produced at an incredibly rapid pace. I will now show you how the scenario promptly adapts as I add each word from the prompt I previously prepared.
Isn’t it astounding?
The images I’m sequentially displaying above are created in real-time, responding to the adjustments I make to the prompt. The fact that each image is generated in just 1 to 2 seconds, showing immediate changes… The power of SDXL Turbo is truly moving. Now, local users who were previously worried about hardware constraints can also produce and practice with more results at a faster pace using the SDXL Turbo model!
Oh, and just so you know, the Generate Forever feature can be stopped by right-clicking it again. I suggest turning it off in situations where it’s not needed to prevent unnecessary operations. Next time, I will guide you on how to use it with ComfyUI!