InvokeAI - SDXL Getting Started

Invoke
29 Jul 202306:40

TLDRThe video introduces the release of version 3.0.1, highlighting the inclusion of SDXL support. It guides users on utilizing the new UI features, emphasizing the use of positive and negative style prompts for effective results. The importance of coherence in the style prompt and avoiding excessive negative prompting is stressed. The video also addresses technical aspects, such as using a fixed VAE for FP16 precision and adjusting image sizes for the SDXL model. Tips on using the refiner for detail enhancement and the recommendation of Euler schedulers are provided. The video concludes with a mention of upcoming SDXL fine tunes and resources for further learning.

Takeaways

  • ๐Ÿš€ The release of 3.0.1 introduces support for the SDXL model, which requires a different approach to prompting.
  • ๐Ÿ“ The linear UI has been updated to include two additional prompt boxes: positive style prompt and negative style prompt.
  • ๐Ÿ”— Using the 'concatenate basic prompt with style' button is recommended for best results when using SDXL, especially for beginners.
  • โš ๏ธ Be cautious with the content of the style prompt to maintain coherence; too much can lead to loss of coherence.
  • ๐Ÿšซ The SDXL model dislikes excessive content in the negative prompt, so focus on specifying what you don't want in a stylistic manner.
  • ๐Ÿ’ก When using FP16 precision, ensure to download a fixed VAE file compatible with the SDXL model to avoid image aberrations.
  • ๐Ÿ–ผ๏ธ For SDXL, start with a 1024x1024 pixel size or similar to ensure coherence with the model's training images.
  • ๐ŸŽจ The refiner model can add fine details but may require some trial and error to achieve desired results.
  • ๐Ÿ”„ The number of steps in the refiner should be based on a 75% completion point of the original steps selected.
  • ๐Ÿ“Š Euler schedulers have been found to be successful in SDXL, while some DPM plus plus schedulers may produce unusual results.
  • ๐Ÿ”ฅ More SDXL fine tunes are anticipated in the coming weeks, with the new Dream Shaper XL model being one of the first releases.

Q & A

  • What is the main feature introduced in version 3.0.1?

    -The main feature introduced in version 3.0.1 is the support for the SDXL model, which includes additional prompt boxes for positive and negative style inputs.

  • How does the linear UI facilitate the use of the SDXL model?

    -The linear UI has been updated to make it easy for users to input their SDXL prompts and use the refiner directly from the interface, by providing a subject and style prompt box that can be concatenated together.

  • What are the recommendations for using the style prompt with SDXL?

    -It is recommended to keep the style prompt concise to maintain coherence and to focus on stylistic elements that the user does not want to see in the output.

  • Why is it important to use a fixed VAE file with SDXL?

    -A fixed VAE file is important for FP16 precision runs to avoid issues with the original VAE file that was broken and caused unwanted dramatic aberrations in the images.

  • What should be the approximate pixel size for the input when using SDXL?

    -The input should have roughly the same number of pixels as the original images that the SDXL model was trained on, such as 1024x1024, to ensure the desired coherence in the output image.

  • How does the refiner model work in conjunction with the base SDXL model?

    -The refiner can add detailed touches to the image, but it is recommended to use it with caution and to expect some trial and error before achieving satisfactory results.

  • What is the recommended setting for the refiner start slider in the UI?

    -The refiner start slider should be set around 0.7 to 0.8, indicating that the base model should end its processing when the image is structurally mostly complete, leaving the finishing touches for the refiner.

  • Which schedulers are recommended for use with the SDXL model?

    -Euler and Euler ancestral schedulers are recommended for use with the SDXL model, as other schedulers may produce undesirable results.

  • How can users get started with SDXL prompting?

    -Users can refer to the SDXL prompt Styles reference created by a contributor on GitHub, or seek more guidance on Discord.

  • What is the significance of the positive and negative style prompts in the SDXL model?

    -The positive style prompt defines the desired style elements, while the negative style prompt specifies what should not be included, allowing for more precise control over the output image.

  • What is the expected development in the near future regarding SDXL?

    -In the coming weeks, it is anticipated that a number of SDXL fine-tuned models will become available, offering users more options and improvements.

Outlines

00:00

๐Ÿš€ Introduction to SDXL Model and UI Features

This paragraph introduces the new 3.0.1 update which includes support for the SDXL model. It highlights the ease of use of the updated linear UI, which now features two additional prompt boxes for positive and negative style prompts. The importance of understanding the SDXL model's different approach to prompting is emphasized, and the recommendation to use the concatenate basic prompt with style button for optimal results is provided. The paragraph also advises on the careful crafting of the style prompt to maintain coherence and avoid excessive negative prompting. Additionally, it addresses the need for a fixed VAE file compatible with FP16 precision to resolve issues from the original SDXL base model and improve image quality.

05:03

๐ŸŽจ Utilizing SDXL Model for Image Creation

The second paragraph delves into practical advice for using the SDXL model effectively. It discusses the importance of maintaining a similar pixel size to the original images the model was trained on, to ensure coherence in the output. Guidance is provided on choosing the right dimensions for the image, with a recommendation to start with 1024 by 1024 pixels. The role of the refiner model is explored, noting its ability to add fine details but also its sensitivity to settings and potential for finickiness. The paragraph advises on the number of steps to use with the refiner and offers insights on selecting the appropriate schedulers, with a preference for Euler or Euler-ancestral. It concludes with an anticipation of more SDXL fine tunes becoming available and encourages users to experiment and seek support from the community.

Mindmap

Keywords

๐Ÿ’กsdxl

sdxl stands for Style Diffusion XL, a model used in the context of the video for image generation. It represents a shift in how users prompt and interact with AI models, requiring a new approach to input prompts. The video discusses the integration of sdxl support in a UI, highlighting its unique features such as positive and negative style prompts.

๐Ÿ’กUI

UI, or User Interface, refers to the system through which users interact with the sdxl model. The video emphasizes the ease of use and intuitive design of the linear UI, which has been updated to accommodate the unique requirements of the sdxl model.

๐Ÿ’กprompts

Prompts are the inputs provided to the AI model to guide the output. In the context of the video, prompts have evolved with the introduction of sdxl, now including subject and style prompts that can be combined for better results.

๐Ÿ’กfp16 Precision

fp16 Precision refers to a method of numerical computation that uses 16 bits to represent floating-point numbers. The video mentions the importance of using a fixed VAE file that supports fp16 for running the sdxl model, as it can affect the quality and coherence of the generated images.

๐Ÿ’กVAE

VAE, or Variational Autoencoder, is a type of neural network used for efficient learning and generation of images. In the video, it is mentioned that a fixed VAE file is recommended for use with the sdxl model to ensure proper image generation and to eliminate unwanted visual artifacts.

๐Ÿ’กimage resolution

Image resolution refers to the dimensions of an image, typically expressed in terms of width and height in pixels. The video emphasizes the importance of maintaining a similar resolution to the original images the sdxl model was trained on, to ensure coherence and quality in the generated images.

๐Ÿ’กrefiner model

The refiner model is a component used to enhance and add details to the base AI-generated image. The video discusses the refiner's ability to add fine details but also notes its sensitivity to settings and potential for producing unexpected results.

๐Ÿ’กEuler schedulers

Euler schedulers are a type of learning rate scheduler used in training neural networks. The video suggests that Euler or Euler ancestral schedulers are recommended for use with the sdxl model, as they have been found to produce more desirable results compared to other schedulers.

๐Ÿ’กSD Excel prompting

SD Excel prompting refers to the technique of using specific styles in prompts for the AI model. The video encourages users to explore resources like the sdxl prompt Styles reference for guidance on effective prompting with the sdxl model.

๐Ÿ’กfine tunes

Fine tunes refer to the process of making minor adjustments to a base AI model to improve its performance or achieve specific outcomes. The video anticipates the release of sdxl fine tunes, which are specialized versions of the model tailored for particular tasks or enhancements.

Highlights

3.0.1 release introduces support for the SDXL model.

The linear UI has been updated to include additional prompt boxes for positive and negative style prompts specific to SDXL.

SDXL requires a shift in how users prompt, with a focus on the style prompt for coherence.

For those unfamiliar with SDXL, using the 'concatenate basic prompt with style' button is recommended.

Overloading the negative style prompt can lead to a loss of coherence.

When using FP16 precision, download a fixed VAE file to avoid issues with the original SDXL base 1.0 model.

A new VAE can be easily added by invoking and placing it into the Auto Import folder.

For SDXL, starting with a 1024x1024 pixel size or similar is recommended for coherence with the model's training images.

The refiner model can be toggled on and off, with the base model often sufficient for fine-tuned outputs.

The refiner adds details but may require fine-tuning with batch settings.

The number of steps in the UI determines the number of steps run using the refiner.

A refiner start value around 0.7 to 0.8 is recommended for optimal results.

Euler and Euler ancestral schedulers are recommended for use with SDXL.

DPM plus plus schedulers may produce undesirable results in SDXL.

SDXL fine-tunes are anticipated to become available in the coming weeks.

The SDXL prompt Styles reference on GitHub provides a useful starting point for SDXL prompting.

The positive and negative style prompts are concatenated together for effective use in the UI.