Learn Generative AI w Stable Diffusion, ComfyUI & GGUF Flux

The new version of Stable Diffusion - SDXL will be available for users from 26 July 2023.  There are two models - the Base model and the refiner model.  Links to these can be found in the resources section for this lecture.

A brief discussion of the practical capabilities and limitations of SDXL 1.0 in rendering images at the 1024 pixel x 1024 pixel scale.

Learn how to use PNG files and JSONs files to create Stable Diffusion XL workflows.  This lecture provides the resources to create Stable Diffusion XL workflows and to test whether your Stable Diffusion XL setup is working properly. The workflow generated by the PNG file may need to be run a couple of times to achieve the render seen in this lecture.

The Stable Diffusion XL workflow will not work with Stable Diffusion 1.5 and SD 2.1 models because it uses newer, SDXL specific nodes.

A brief quiz and thinking section on SDXL workflow.

Answers to the Kitten on the Hill Quiz and some ideas on improving the template used in that video.  Resources can be found in the tutorial "SDXL - First Steps with Stable Diffusion XL Workflows"

Advice on improving experience of downloading large files especially on slower internet connections.

ComfyUI has many powerful extensions that you can use to extend, organize, personalize and simplify the software to your hearts content.  The powerful and versatile ComfyUI manager is presented here.

The lecture demonstrates how to install a pinned version of the manager for use during the remainder of the course.

A lightning tour of the capabilities of the ComfyUI Manager.

After this lecture you will understand how to use the powerful WAS node extension to access presets or styles from Automatic1111 - the most popular version of Stable Diffusion.  The lecture provides some practical insights into how to develop a strategy for choosing extensions for working with ComfyUI.

Users can take advantage of the coupon FREEAUTOMATIC to access the Pixovert Automatic1111 Stable Diffusion Course at a discount throughout August 2023: at https://bit.ly/FREEAUTOMATIC (or use the resource link for this lecture).

This is an important update on the Efficiency Nodes which are covered in upcoming lectures.

Learn how to maximize efficieny through the use of Efficiency Nodes and along the way, how to make the ComfyUI UI, even more Comfy.  The Efficiency Nodes extension allows us to add quite a bit of crucial functionality that is built-in in the Automatic1111 version of Stable Diffusion.

Learn to modify images using the classifier-free guidance and Clip Set Last Layer (clip-skipping).

This lecture provides guidance on advanced techniques in prompting.  This is a very technical lecture that benefits from close attention and hands-on practise. 

The json files provided will launch two workflows - one for the beginning state and the other for the post edited state (E).  These workflows depend on several extensions.  The workflows may need to be adjusted by students if the extensions are not working properly.  I suggest trying to update ComfyUI and the extensions (use the Manager extension for this) as a first step if you see any errors.  Stable Diffusion's discord may be another useful resource if you have problems with your installation.

Generative Fill is Adobe's name for the capability to use AI in photoshop to edit an image. Something that is also possible right in ComfyUI it seems.  This is an introduction to what is in fact a very powerful set of features in ComfyUI.  More powerful methods are featured in the more advanced courses.

The new ComfyUI offers significant improvements over the older interface.  This video covers some of the key recent changes and how you can customize Comfy for maximum information or to calm down the informative popups.

What this video covers
This video explores the Flux Krea unified model family, including why these models are so effective for portraits, subtle emotional prompting, and distinctive color-rich imagery.

Why it matters
These models are educationally valuable because they are powerful but not effortless, which makes them especially useful for understanding prompting limits, CFG behavior, and model-specific workflow decisions.

Who should watch
This is for ComfyUI users and generative AI creators who want better model judgment for portrait, landscape, and substitute cloud workflows.

This lecture will introduce you to state-of-the-art practices in compressing and optimizing AI visual models, focusing on the FLUX, KREA, and Crayon models within ComfyUI workflows. You will learn the technical foundations and choices behind quantization using GGUF format, and how these choices affect the deployment and quality of AI-generated images. By the end of the lecture, you will:

• Understand the process and benefits of quantization for reducing model size and hardware requirements.

• Identify optimal GGUF variants and encoder configurations for your GPU capabilities.

• Compare quality and precision tradeoffs between different model variants.

• Deploy and integrate FLUX and KREA models in ComfyUI, troubleshooting node workflows and encoder compatibility.
  
  FLUX: An AI model for generating high-quality images with advanced output control.

  Krea: A creative model for image generation, used with FLUX to achieve varied styles and enhanced results.

  Quantization: Reducing the numerical precision of a model (e.g., from 16-bit to 4/8-bit weights) which decreases file size and speeds up inference, with some quality trade-offs.

  GGUF: A modern file format for quantized model weights. It provides more efficient storage and lower resource consumption for running models locally.
  

This lecture provides a comprehensive exploration of portrait generation using the FLUX KREA GGUF model in ComfyUI. Students will learn how prompt structure and style suggestions impact the output of AI-generated portraits.

A copy of the main workflow is attached in PNG format.  It is the second stage of the style evolutions.

This lecture guides students through the use of the FLUX KREA model for generating landscapes and still life images with AI in ComfyUI. Core concepts include the capacity of the model for artistic styles like landscapes, still life, and balanced compositions. Students will examine, aspect ratio choices, the impact of step count on detail, and the plausibility of outputs. By the end of the session, participants will:

• Understand how to leverage capability of the T5 transformer and the Krea model to influence style and composition in non-portrait scenes.

This lecture focuses on creative prompt engineering when using the FLUX KREA model for stylized and fantasy AI-generated images. Students will examine the model's strengths in portraying artistic, fantasy, and expressive works using clear, ordinary English prompts, and learn about common quirks (like unexpected signatures) that can arise.  Key skills include adjusting prompt complexity, and appreciating the unique output characteristics compared to older AI systems. By the end of this session, students will:

• Understand how prompt phrasing and detail, and sampling depth influence the style and plausibility of generated images.

• Experiment with transforming basic ideas into creative, expressive visual outputs (e.g., "a bale of hay in a vibrant expressionist oil painting").

• Recognize limitations, such as persistent signatures in art-style outputs, and discuss possible sources for such artifacts.

• Appreciate the ease and power of using natural language for creative control with FLUX KREA.