Two popular Stable Diffusion tools, that could not be more different.
Last summer I posted an article on 3 Stable Diffusion tools to run on your PC, supporting Intel Arc GPUs. And as things go with AI, so much has advanced. Thankfully, Intel software frameworks and extensions have not only allowed for latest features like fast LCM LoRas to run on Intel Arc GPUs, we also have support across more GenAI tools than we did 6 months ago.
In this article I will cover two additional webui’s for Stable Diffusion. Both more powerful than what we have seen before and depending on the type of user you are, one of these is probably the right choice for you.
But these new tools could not be more different. They sit at the ends of the spectrum when it comes to complexity and user experience. Fooocus being a simple interface where configuration changes are often done for you vs ComfyUI with the power of a node system allowing for endless configurations and custom workflows.
Take a look for yourself in the video below where I walk you through using each interface, and read on to get details and features of each tool. One of these may be the right tool for you, and both run great using discrete Intel Arc GPUs. Installation and support details linked below.
Fooocus WebUI: The Fooocus webui has to be the most interesting blend of advanced image diffusion features while at the same time being the a simple toned down interface. Fooocus is set apart by automating many steps a user would otherwise do manually. As new models, refiners, and other techniques are added to generative AI, Fooocus works to make the work easier. For example Fooocus automatically comes with the JuggernaughtXL model, which is a great multipurpose model for generating realistic, cinematic and concept art like images. As well Fooocus automates features like adding a refiner so the output is cleaner and more detailed. And when switching between SDXL and SDXL with LCM, you only click one button. In most other UIs, you are making at least 3-5 configuration changes when you make that switch. Lastly Fooocus has one of the better image input features of any tool, especially for outpainting, where you can extend an image in any direction adding content that was not in the original image.
Bottom line, if you want something user friendly, highly intuitive to use, while allowing for powerful features and customization as you need it, this is the webui for you.
Stand out features
- Simple and clean interface, that doesn’t overwhelm and is very intuitive.
- SDXL output. By default all images generation is high res.
- Intelligent configuration where the best and fool proof configuration options are done for you.
- Installs with the Juggernaut SDXL model. (no need to download from Hugging Face of CivitAI)
- Powerful and easy to use outpainting and inpainting features.
- Simple check box for 3 speeds / quality of image generation: Fast (20 samples), Quality (60 samples), Extremely Fast (8 samples LCM Fast LoRa)
- Style options that direct the output to specific looks and styles
- Hidden but accessible advanced and developer features that allow you to override the defaults, allowing you to fine tune the output
Installation Instructions and Discussion Thread on Intel Insiders Discord (for discrete GPU only)
ComfyUI: At the other end of the spectrum is the increasingly popular ComfyUI tool for image generation. If you watch a lot of Stable Diffusion videos like me, I’ve seen many YouTubers shifting from A1111 to ComfyUI as it supports a deeper set of customizations with custom nodes from the community. This tool is perfect for those who like to tinker with settings and push the boundaries of what current GenAI technology can do. It’s a powerful webui and we are just scratching the surface on what this tool can do for you.
What sets ComfyUI apart from other tools is its use of a visual nodes system to define how you want to generate your image or video. If you are familiar with Unreal Engine, Blender and Unity, node systems are a visual programming method which does not require coding. This is done by connecting the output of one node function and plug it into the input of another node function. Outputs can be connected multiple nodes that do different things creating a flow of processes. In Comfy there are nodes for things like; image input, text input, model input, LoRa encoding, ControlNet preprocessors, and so on.
The Community also creates and distributes custom nodes allowing for a plug-in like architecture for other applications outside ComfyUI to operate with this node system to add functionality to image generation. Custom nodes may allow for web cams or applications such as image editors with filters and to influence the image processing.
Additionally, node workflows can be saved and reloaded for the future, or distributed as JSON images, allowing you to drag and drop visual workflows directly into your UI space for fast loading. This UI is quickly becoming an extensible development platform that doesn’t require any coding skills.
Stand out features.
- Complex and powerful node based system
- Broad and deep set of configurable controls and features without having to code
- SD 1.5 and SDXL support, with support of custom; checkpoints, LoRas, VAEs, and refiners, and more.
- Growing ecosystem of node creators for 3rd party app integration such as; advanced tool manager, processor and memory utilization, video and web stream input, and more
- Supports ControlNet, In painting/outpainting, fast LCM LoRa, Animate Diff, SD Video, face swap
Installation Instructions and Discussion Thread on Intel Insiders Discord (for discrete GPU only)
Watch the video above to see these tools in action running on an Intel Arc GPU and Follow these below for further information in Intel Arc and AI on Intel Hardware. For continued content and engagement join the Intel Insiders Discord Community where I and others from Intel regular converse with the community.
