What is SD.Next

SD.Next is a Stable Diffusion interface based on the Automatic1111 code base. Compared to the original Automatic1111, the changes it has received this year allow it to support different layouts and color schemes, give it improved compatibility with different systems and graphics cards, and give it extended support for models in different formats. SD.Next also comes with a number of plug-ins included. Read the list of the features on its GitHub page.

There are installation instructions on the SD.Next GitHub page, but if you'd like to follow along with a video, you can watch this video on YouTube.

SD.Next is supposed to run on more types of systems than Automatic1111. If you are using an AMD GPU, check out this video about installing SD.Next on AMD.

SD.Next does has some nice features for people just starting out, such as visual browser to choose online models that you can click to download and install. It also has a floating help window that shows info about features as you hover your mouse over them.

If you've already got a lot of models on your system (to use with Automatic1111, ComfyUI, or Fooocus), then you don't need to download new models for SD.Next. Under the Systems tab > Settings > System Paths, you can change where SD.Next will look for models. If you paste the full paths of your existing models folders into these slots, you can save yourself a lot of hard drive space.

When it first starts in the "Standard" theme, the SD.Next interface seems similar to the original Automatic1111, with only a few things moved around. There are small changes such as not having a "PNG Info" tab for your history. The function of dragging an image in to inspect and re-use the prompt and settings is still present, though, in the Process tab. Also, adding a LoRA is hidden under a box labelled "Networks."

Changing the interface from "Standard" to "Modern" rearranges things to such an extent that even experienced Automatic1111 users could get completely lost. When I tried that setting, it actually took me a while to even find the preferences again in order to change it back. Some people might like the new interface layout better, but I don't see a lot of community support for it yet. When I searched YouTube, there were very few SD.Next tutorials, with none on adapting to the Modern interface.

SD.Next has received some changes that are billed as performance improvements. However, when I switch between models, generate images, resized them, or use image-to-image, I couldn't detect any speed increases. In fact, my overall impression is that SD.Next seemed slower than Automatic1111, and is clearly slower than ComfyUI or StableSwarm.

It does have some nice features and options in the interface, such as HiDiffusion (which seems similar to what's also known as Koyha Hires. Fix, allowing you to directly generate images at much higher than a model's base resolution, without the doubling artifacts you'd usually see, and without needing the image to be resampled in multiple passes.) Another nice feature is a library of built-in styles that SD.Next can apply to prompts, similar to the style library that Fooocus provides.

The latest version of SD.Next supports Flux, but it supports Flux in a limited way, loading diffusers models derived from the original Flux.1-Dev. When it gets updated to use the full, original Flux.1-Dev model, I'll do a comparison on speed and performance running it against Forge WebUI and SwarmUI. For now, it seems too early to say whether SD.Next will become a great way to run Flux.

When you come to a fork in the road, the road less travelled could still be a road worth taking. If you love some of the new features, want to enjoy the look of a new interface, or you are using an AMD card and want something like Automatic1111, but you can't get Automatic1111 running on your machine, then SD.Next could be for you.