A simple comparison using SDXL model by Automatic1111 and ComfyUI-Before you start

Recently, the Stability AI team unveiled SDXL 1.0, an open model representing the next step in the evolution of text-to-image generation models. It's fully compatible with Automatic1111 and works seamlessly on ComfyUI as well. Currently, ComfyUI has become the second most popular app in the Stable Diffusion Ecosystem. However, there are few videos discussing the differences between these two applications. Here, we conducted tests on both systems to provide you with a clear and simple understanding of each. This way, you can gain more insights into these systems before delving into text-to-image generation.

The ComfyUI system allows easy modification of settings, eliminating the need to download any extra models like LoRA or ControlNet. You can effortlessly use these models in both systems.

While WebUI offers a straightforward way to use the SDXL model, it can be a bit slow, especially with older graphic cards. On the other hand, ComfyUI can generate images at a much faster pace. However, for beginners, it might be slightly more challenging to handle. Fortunately, you can use the PNG file containing almost all the necessary information to generate the image, significantly accelerating the learning process.

Both applications excel at producing stunning and realistic images. Using only a few keywords, the generated image quality is truly astonishing. SDXL is a game-changer in the field of text-to-image generation.

The EXCEL result is taken using
set COMMANDLINE_ARGS=--xformers --no-half-vae
set COMMANDLINE_ARGS=--xformers --medvram --no-half-vae
The image generation speed increased dramatically. Thanks for Mr. Enricii comment. He pointed out it may help. In fact, it is really a lifesaver. I will make a corrected version soon.

📚 RESOURCES from the Internet

Stability AI news

SDXL 1.0 base model on Huggingface

SDXL 1.0 refiner model on Huggingface

SDXL knowhow-workflow of the best way to test this model, a game changer

ComfyUI on GitHub

7-zip open source compress software

OSS Insight (Open Source Software Insight)

ComfyUI workflow file

Subscribe to my channel

Twitter

Facebook

#TubeUnderdeveloped #Howto #3dPrinting #AICorner

[Enricii]

Hello! I've got a RTX 3070Ti with 8 Gb VRAM and I can generate 1024x1024 images in A1111 in 30-50 seconds (without refiner). In ComfyUI I can generate more or less in the same time, but with refiner. I would say the difference is there, but not as big as yours.

Have you ever tried to use commands --medvram, --no-half-vae in the BAT file before launching A1111? That should help.

[JeanKulacek-ngdg]

thanks for allowing us to get your Comfyui workflow it will me to study and improve mine. Thanks again

[filosofiahoy]

The girl with the flowers looks much better in Automatic1111 than Comfy in Comfy the image was flattened into 2D.

[demoran]

Controlnet is incompatible with SDXL. Stability intends to release a replacement for it, but no ETA.

[filosofiahoy]

You need to compare SDXL1.0 VERSUS EpiCRealism - Puere Evolution V4 Side to side, Hands, tubes, Iris, very complex prompts etcetera-

[globalmodelsdotbiz]

Correct, ComfyUI blows Automatic1111 out of the water.