Fooocus AI tried AI image generation has never been easier!

You probably know this: You hear about all these cool AI image generators, you want to try it out, and then..?

Then you spend three hours installing the correct Python version, google any wild error messages and maybe even give up frustrated at the end. In any case, it was the same with me when I first wanted to try Stable Diffusion.

But at some point I'm over Fooocus AI Stumbled, and frankly: I would have liked to have discovered this earlier. It took less than five minutes from the download to the first image generated. No kidding. And the results? Just as good as the more complicated tools, but without all the frustration.

Let me show you why I only use Fooocus now and why this might be interesting for you as well.

What is Fooocus?

Okay, short classification: Fooocus is basically an AI image generation software based on the same SDXL model as all the other tools – Automatic1111, ComfyUI and as they are all called. The difference? The developers asked themselves: “What if we just... just did it?”

They really sat down and said: ‘Maximum three clicks from download to first image.’ And you know what? They've done it. No hours of watching tutorials, no command line acrobatics, just start and get started.

The thing is kind of like Midjourney, except it runs on your own computer and is completely free. And compared to the other SDXL tools, Fooocus automatically does all the annoying technical work you would otherwise have to do yourself.

Installation: It's never been this easy

Normally I'd write a novel about system requirements now, but honestly – if you've bought a halfway up-to-date PC with an Nvidia graphics card in the last few years, it's probably running. Even with only 4GB of VRAM, it works thanks to some Microsoft magic (virtual swap). I tested it on my old GTX 1660 Ti – it works fine.

Here's how to get started

Watch out, I'll show you the installation now, and you won't believe how simple it is:

Go to GitHub lllyasviel/Fooocus (but be careful, sometimes fake websites appear on Google – go directly to GitHub). Download the latest version, currently this is the 2.5er. This is a ZIP file, so 2-3 GB in size.

Unpack them somewhere – I just threw them into a folder ‘KI-Kram’ on the hard drive. Then go to the folder and double click on run.bat. That was’s. Seriously.

At the first start, the program automatically downloads all the models it needs. Depending on the internet, this takes a few minutes. Get some coffee, and when you get back, it's all done. The browser opens automatically with the Fooocus interface, usually under localhost:7865.

I have a screenshot here... oh no, you can imagine: There is a large text box at the top and a ‘generate’ button at the bottom. Nothing more is needed at first.

Seriously, that was the whole installation.

At first I didn't think it could be that easy. I waited 20 minutes for something complicated to happen. But no, that was it. No Python installation, no virtual environments, no dependencies hell. Just double click and you're done.

In comparison: When I installed Automatic1111 last year, I sat on it for two evenings. First installed Python in the correct version, then cloned Git, then tried some Torch versions because something was constantly not working. ComfyUI was even worse – I gave up at some point and stayed with A1111.

And then Fooocus comes along and turns the whole drama into a five-minute process. Why haven't others done it this way long ago?

My first test: "Wow, that was fast"

Okay, so this thing's going on now. I took the classic test: ‘a cat sitting on a windowsill’ and clicked on Generate. No more settings, just the prompt.

Fifteen seconds later, I had a picture. And not a shitty picture, but one that looked like I'd been screwing around the settings for hours. The cat looked realistic, the fur had details, the lighting was right – everything was just fine.

I thought it was lucky, so I tried again. ‘An elderly wizard in a mystical forest’ – again such a generic prompt as can be found everywhere as an example. Another 15 seconds, another big result. The beard had individual hair, the robe had wrinkles and textures, in the background were fireflies, which I had not explicitly mentioned.

The difference to other tools: Automatic vs. manual work

This is where it gets interesting, because that is where Fooocus really stands out from the competition. Let me show you what usually happens when you work with SDXL:

The normal way (with Automatic1111 or ComfyUI)

Imagine you want to generate an image. Here's what you usually have to do:

First of all, you have to get the right models. That means you go to any website such as Hugging Face or Civitai, look for the SDXL Base Model (there are different versions – which one was the best again?), download it. This is often 6-7 GB per model.

Then you need the Refiner model. This is such an additional model that improves the details. So again 6 GB download. But wait, where does this have to go now? Yes, in the models folder, but in which subfolder? Was that checkpoints or models/stable diffusion?

Then you have to see if you need a UAE (no idea what that is, but supposedly important). You also need to put it in the right folder. If you forget that, your pictures look funny.

When you finally have everything in the right place, it's time to generate. But wait – first you have to choose the right model in the GUI. Then activate the Refiner. Then set the step at which the refiner should start. Then choose the sampling method. Then set the steps. Then the Guidance Scale...

You see, this is work. And if you don't know exactly what you're doing, the results are often meh. I used to spend hours spinning around all these parameters.

The Fooocus path

Start the program. Promptly enter. Click on Generate. I'm done.

Really, now, that's all. But what happens in the background? Fooocus does exactly what I just described – just automatically and correctly:

Model management: At the first start, Fooocus automatically downloads the best SDXL models. Not just any, but the ones that work well. They also automatically end up in the right folder. You don't have to worry about it.

Refiner integration: It's going to be technically exciting here, even if you don't have to see it all. Fooocus has a trick: They use the refiner within the same sampler as the base model. This sounds like nerd talk at first, but the effect is real – the images look more coherent because the refiner can build on the information from the base model. With A1111 and ComfyUI, they run in separate samplers, which does not have this advantage.

But do you know what's best? You don't have to know all this. It just works.

Prompt optimization: This is my favorite feature. Fooocus has a GPT-2 model running internally (no idea why they're taking GPT-2, but no matter) that automatically expands and improves your prompts.

You write ‘a cat’ – Fooocus turns it into ‘a photorealistic cat with detailed für texture, professional photography, high quality, sharp focus, natural lighting, depth of field’. You don't have to write it yourself. The system does this for you based on what experience has shown to work well.

With other tools, you have to put all this into the prompt yourself. I used to have huge text files with ‘good prompts’, which I then always copied and adapted. With Fooocus? No longer necessary.

The surface: Simple but not stupid

When you open Fooocus for the first time, you may think: ‘Hm, that looks pretty... empty.’ That's also true. Above is a large text box for your prompt, including one for the negative prompt (things that should NOT be in the picture), and then there are a few basic settings.

You can choose the aspect ratio (1:1 for Instagram, 16:9 for YouTube thumbnails, etc.), set how many images to generate, and choose from different style presets. That was’s first time.

I found it almost too simple at first. Where are all the settings? But then I realized: I don't need them. For 95% The basic settings are sufficient for my pictures.

For the nerds: Advanced mode is also available

If you want more control, no problem. There is an advanced tab where you can set all the technical stuff. Performance modes (Quality, Speed, Extreme Speed), custom resolutions, Guidance Scale, Sampling Steps, and so on.

The difference with A1111 is: These settings are optional. The program works great without them. With A1111, the GUI will blow your ears at the first launch with a hundred different options, and you don't know where to start. At Fooocus, they're there when you need them, but they're not annoying when you don't need them.

I've now realized that I almost never touch the advanced settings. The auto settings are just good enough.

GUI vs. command line: Why Fooocus scores here

Have you ever tried working with the command line? So that's right? With any commands you have to type in?

With many SDXL setups, you can't get around it. Even if the tool has a GUI, you have to start it first from the command line. And if something doesn't work, it says: ‘Open the command line and enter the following commands...’

I'm a developer, I can handle the command line. But I totally understand when people say: ‘No, let's face it, that's too complicated for me.’ My friend also wanted to generate images the other day, and when I wanted to explain to her how to start A1111 (‘So, you open the command line, go to the right folder with cd, “activate the virtual environment with...”), she rolled her eyes only.

With Fooocus? ‘Just double-click run.bat.’ Everyone understands that. A browser tab will open and go. No cryptic commands, no ‘activate venv’ or ‘pip install’ or whatever.

This is the point where Fooocus is better not only for beginners, but also for people who just want to generate images without messing with IT stuff.

Styles: 275 ready-made presets that really make a difference

I have to admit, when I saw that Fooocus has over 275 different style presets, I thought: "Oh God, more options I don't need." But then I tried them, and they are really good.

A style isn't just a few additional prompt words. These are complete configurations that also optimize the technical parameters. If you choose ‘Cinematic’, you not only get cinematic looks, but the system also adjusts the sampling parameters to get the best result for this style.

A few examples that I found particularly cool:

cinematic: Perfect for anything that looks like a screenshot from a movie. Cool lens flares, dramatic lighting, the full program. I took some pictures for a YouTube thumbnail – it looked really professional.

anime: If you're into anime style, that's the way to go. No weird hybrid results, but really clean anime look. My buddy is doing character designs for his indie game.

Photographic: This is my default style. Looks like someone took pictures with a really good camera. Not as artificial as some AI images.

You can also combine several styles. Cinematic + Fantasy, for example, gives you cinematic fantasy scenes. The system is clever enough to mix the two styles sensibly.

The horny thing about it: With other tools, you'd have to craft these effects all by yourself. Test different prompts, adjust parameters, maybe load LoRAs. At Fooocus, you just click on a style and it works.

Practical examples: How I Really Use Fooocus

Enough theory, let me show you how I use this thing in real life:

Content for Social Media

From time to time I do content for Instagram and LinkedIn. I used to look at Unsplash or Pexels for stock photos that never fit exactly. Now I'm doing it like this:

I think what I need – for example ‘modern workspace with laptop and coffee cup, morning light’. Choose the "Photographic" style, set the Aspect Ratio to 1:1 for Instagram (or 16:9 for LinkedIn), and generate 4-5 variants. It won't take two minutes.

Then I'll pick the best one, maybe I'll do some color grading in Lightroom (I don't have to, but I like it) and the post is ready. From the idea to the finished picture: 10 minutes max.

This would have taken longer with A1111, because I would have to go through all settings first. With Fooocus? Quickly in, choose style, done.

Conceptual art for a project

The other day I had a project where I wanted to visualize different room concepts. Such as ‘futuristic office space with plants, natural lighting, minimalist design’.

With Fooocus I first generated 20-30 different variants, just to see what is possible. It may have cost me half an hour. Then I picked out the best three and used them as image prompts (more on that) to create even more specific variants.

In the end, I had a really good collection of concepts that I could then continue working with. If I had done this with a designer, we would have ended up with a few thousand euros and several weeks in terms of cost and time. With Fooocus? One afternoon and 0 Euro.

Illustrations for a website / printed matter

For websites and printed matter, I need pictures on a regular basis. Header images, illustrations for certain concepts, such things. In the past, I really spent time finding suitable stock photos and editing them.

Now I'm just writing a prompt based on the article topic. For example, for an article on productivity: ‘person working focused at desk, plants in background, warm lighting, productive atmosphere’. ‘Digital Art’ or ‘Photographic’ style, whichever is better.

The best: The pictures are unique. No one else has the same pictures. No more ‘have I seen it anywhere else’ moment.

Image prompting: From good to awesome

Okay, now it's gonna be really cool. Fooocus can not only text-to-picture, but also image-to-picture. You upload an image, and Fooocus generates variations of it or uses it as inspiration.

I used this the other day: Had a photo of a landscape, the composition found good, but the weather was meh. So I uploaded the photo and wrote: ‘same composition, but golden hour lighting with dramatic clouds’. Boom, perfect result.

You can also control the influence of the reference image. At 100% remains very close to the original, at 30% takes’s only rough inspiration. And you can even combine several images – composition of picture 1, color style of picture 2, details of picture 3.

This works easier with Fooocus than with the competition, because the entire technical parameters are already sensibly preset. On A1111 you would have to play around with Denoising Strength, ControlNet and what do I know. At Fooocus, you upload the image, and it just works.

What Fooocus is not

To be fair, Fooocus is not the best solution for everyone. There are situations where the other tools fit better:

If you really want to control every single aspect of the generation process – every node, every parameter, every step – you are better served with ComfyUI. Fooocus automates a lot, and sometimes that means less control.

If you have very special custom workflows with tens of different ControlNets and multi-stage processing and what do I know, then A1111 or ComfyUI offer more possibilities.

And if you want to work with Stable Diffusion 1.5 or other older models (no idea why, but some people like it), then Fooocus is not optimal. This is really trimmed to SDXL.

For 90% The user is Fooocus but the better choice. Unless you really want to go deep and control each parameter yourself.

Performance: Is it also running on my computer?

I often get the question: ‘Is my GPU good for this?’ Short answer: Probably yes.

I have tested it on different systems:

Gaming PC (RTX 4060TI 16GB + 32GB RAM): Runs like lubricated. 1024×1024 images in 12-15 seconds. Can also make higher resolutions with upscaling, no problem.

Old Laptop (GTX 1660 Ti, 8GB RAM): It works too. Takes longer, so 25-30 seconds for a picture, but it works. Sometimes I have to switch to ‘speed’ mode instead of ‘quality’, but the results are still good.

Alternative computer (RTX 2070 Super, 8GB VRAM): Runs smoothly. It even has only 8GB of RAM, but Microsoft’s Virtual Swap option still works.

In comparison, Fooocus is even slightly faster than A1111 with the same settings. It's probably because the pipeline is more streamlined. But the real gain is not the pure speed – it is the time you DO NOT spend optimizing settings.

If it doesn't work: Troubleshooting

Normally, Fooocus simply runs. But sometimes there are problems. Here are a few things I've encountered:

Out of Memory Errors: Sometimes it happens when too many other programs are running. Solution: Other GPU-hungry programs close (Chrome with 50 tabs, for example), or activate the ‘Low VRAM Mode’ in the settings. It usually helps.

Generation takes forever: Check first whether the ‘Quality’ mode with max Steps is not inadvertently active. ‘Speed’ or ‘Extreme Speed’ are perfectly fine for testing. Upscaling also takes time – if you don't need it, it deactivates.

Pictures look funny: Mostly it is at the prompt. Be more specific. Or add things to the negative prompt that you don't want. ‘blurry, distorted, bad anatomy’ often helps.

Program does not start: Check firewall or antivirus. Sometimes they block the initial download of the models. Or the port 7865 is already occupied – then you have to change it in the config file, but that was never necessary for me.

Honestly, I had fewer problems with Fooocus than with any other tool in this area. Most things just work out of the box.

Updates and maintenance: Runs by itself

Fooocus is being actively developed. Every few weeks there are new versions with bug fixes and features. Updating is super easy:

Either you just download the new version and unpack it into a new folder. If you have custom models, copy them over, done.

Or, if you have Git on it, go to the Fooocus folder and make git pull. Even easier.

The development is constantly progressing. Currently version 2.5.x is on, and it brings really cool improvements in the inpainting (i.e. editing parts of images). I'm looking forward to what's next.

My conclusion after a month of intensive use

I've been using Fooocus here and there for some time now, and I can honestly say: I don't miss the A1111. Not a little.

The pictures are just as good – of course SDXL is. But the way there is so much more pleasant. I open the program, type a prompt, and 20 seconds later I have a good picture. No half hour settings adjustment, no troubleshooting, no frustration.

For content creators, designers, or just people who want to make cool pictures, Fooocus is perfect. It takes away the technical complexity from you without patronizing you. You can still go deep if you want – but you don’t have to.

Three things I like best:

Firstly,: The installation. I can recommend any Fooocus without having to send tutorial videos for an evening. “Download it, double-click it, get it done” – everyone understands that.

Second,: Automatic model management. I no longer have to worry about finding, downloading and organizing the right models. Fooocus does that. And honestly, that's a huge time saver.

Thirdly,: Consistency of results. On A1111 I always had those days where suddenly all the pictures looked like shit and I didn't know why. With Fooocus? It just works. Always. The car settings are so well tuned that you can rely on them.

So, now it's your turn!

If you have read so far, you now know more about Fooocus than most people who already use it. Time to try it yourself!

Here is the plan: Go to GitHub, search for ‘lllyasviel/Fooocus‘, download the latest version (v2.5.5). Unpack the thing, double-click run.bat (or launch.py if you're on Linux/Mac), wait a few minutes while the models load, and then get started.

Start with something simple: ‘a beautiful landscape’ or ‘a cute cat’ - you know the classics. Look at what's happening. Try different styles. Experiment with image prompting. Have fun with it!

Happy generating!