What are XL models?
A reminder: AI models are like the engines behind your AI generation. Each model is created using their own unique dataset. They each have their own style and capability and need to be used with slightly different settings for the best result.
XL models: Bigger and Better
XL models on PixAI use SDXL as the base, which is an improved version of the previous SD models. For your generations on PixAI, the most important differences to note are that XL models give you:
Larger images with more pixels
Higher resolution
Better anatomical accuracy
Better prompt adherence
Additionally, here are some notes to keep in mind:
XL models don't work with regular LoRAs, XL models only work with XL LoRAs.
XL models start at a larger image size and cost more credits
XL models are currently not compatible with advanced features such as a selection of ControlNet methods, Composition, and Outpainting
Why should I use XL models?
Ever get frustrated with murky details in your images? Wonky fingers? The AI not understanding complex visual descriptions like "seen through the view monitor of a camera", or "holding a pair of chopsticks"? Do you ever run into issues with enhancing your image and having everything distorted as a result?
XL models are better at fixing all of these issues. They are built for larger images, have better human anatomy and understand difficult prompts of complex composition and poses.
And if for some reason you don't like the aesthetics of the XL models, you can always try creating a base image with XL and using it as a reference image to generate again with a regular model.
Why are XL models more expensive?
XL models don't really work with smaller sizes such as 512x512 and 512x768 pixels, so as a result, your smallest image size for XL models start at 768x1280 pixels, and that will cost more credits.
How do I generate with XL models?
Change the way you write prompts!
Your old prompts and negative prompts that work on regular models might not work very well with XL models. We will provide a more detailed guide on how to prompt with XL models at a later date. In the meantime, we suggest reading the model description of specific XL models carefully and learning from the prompts written by other creators using XL models.
Generate with XL LoRAs
XL models only work with XL LoRAs. Regular models only work with regular LoRAs. You cannot mix them together. The current selection of XL LoRAs might be a little limited, but we have just released the XL Model training feature on PixAI, so if you don't find an XL LoRA you like, you can create one yourself.
Generating with Advanced features
XL models are currently not compatible with advanced features such as a selection of ControlNet methods, Composition, and Outpainting
How do I train XL LoRAs?
A reminder: LoRAs (Low-Rank Adaptation) are like add-on modifiers for AI models. You can add loras while you generate to teach the AI to do a specific thing, such as a character, an art style, a pose, or an aesthetic.
XL LoRAs:
Before you start, please note that XL LoRAs are only compatible with XL models, you will not be able to use it with a regular model.
Once you're sure that you are using this LoRA for an SDXL model, you can specify your model type as SDXL with this option.
Prepare your dataset:
Since XL models are trained on larger images, you also need to train your XL LoRA on larger sized images for it to work.
Upload Images at least 768 pixels in size; smaller images will produce poor results
Make sure your images are clear and in-focus