ComfyUI Tutorial Series: Ep08 - Flux 1: Schnell and Dev Installation Guide

The speaker introduces episode eight of a tutorial series focusing on Flux, a new AI model developed by Black Forest Labs. Flux is favored for its performance and is available in different versions: Pro, Dev, and Schnell, with the Pro version accessible only through API. The tutorial will guide viewers on installing and using these versions, comparing their quality and speed. The Dev version offers high quality but requires more computational resources, while the Schnell version is faster but of lower quality. The video also discusses the non-commercial Dev license and the Apache 2.0 license for Schnell, clarifying their usage rights. The speaker provides practical steps for downloading and setting up the models in Comfy UI, including the floating-point 8 versions for easier use.

The tutorial continues with a detailed walkthrough on setting up the Flux models in Comfy UI. The speaker explains how to adjust the workflow settings for the Schnell and Dev versions, including the use of an empty latent image, the recommended K sampler, and the necessary steps for the Oiler sampler. The video demonstrates the process of generating images with Flux, highlighting the differences in processing time and image quality between the Schnell and Dev versions. The speaker also shares tips on optimizing the workflow, such as using the Flux Guidance node and experimenting with different values for better results. Additionally, the integration of Styles into Flux is explored, showing how to combine prompts and styles to influence the generated images.

This section delves into Flux's ability to generate various artistic styles and the optimization of the Comfy UI workflow for beginners. The speaker tests different styles like biomechanical art and portrait cinematic, noting Flux's capacity to understand and apply these styles. A more compact workflow is created for ease of use, with instructions on hiding unused nodes and adjusting settings like width, height, and the order of style importance. The video also addresses the issue of blurred images in Flux, offering a solution by changing the sampler and scheduler. The speaker shares their experiences with Flux's performance on different prompts, emphasizing its strengths in rendering realistic images and handling details like hands effectively.

The speaker discusses the setup for the regular version of Flux, which requires additional CLIP models and a more complex workflow involving the VAE file and the unet folder. The video provides a step-by-step guide on downloading and placing the necessary files, and testing the model in Comfy UI. A comparison between the floating-point 8 and the regular version of Flux is made, noting the differences in detail and generation time. The speaker also introduces new, faster versions of Flux for users with lower VRAM, explaining the installation process for the nf4 versions of Flux and demonstrating their performance in Comfy UI.

This part of the tutorial focuses on evaluating the performance of different Flux models and enhancing the realism of generated images. The speaker compares the floating-point 8 and nf4 versions of Flux, discussing the impact of file size on image quality and generation speed. The video also introduces the use of a realism-enhancing Laura model for Flux, demonstrating how it can improve the realism of portraits and other images. The speaker concludes with a summary of the time taken for each model to generate images and provides organized workflows for download on their Discord channel, encouraging viewers to subscribe and engage with the content.