PlayAI DrawingNow I don't have to worry about beingOld HuangThe knife technique is "backstabbed"! All it takes is a piece of 4 years oldGTX 1650(4GB video memory), AI image output is comparable to the best open source model SDXL.GitHubNew projects that have topped the list for several consecutive daysFooocus, the star rating exceeded 4k in three days,ControlNetauthorProduced.

GitHub project address:https://github.com/lllyasviel/Fooocus

Before it came out, everyone wanted to run smoothly.Stable DiffusionThe latest XL model must be equipped with a 4060ti with 16G video memory (the native 3060 with 12G video memory is very difficult to run).

Unlike other open source AI tools, Fooocus "focuses on generation itself", not only has low hardware requirements, but is also easy to use.Very friendly——

No parameters need to be adjusted during the whole process. Just click the mouse.Generate an image in 3 steps.

△Photo source Twitter @Photogenic Weekend

Some netizens said, "This is simply the culmination of Stable Diffusion and Midjourney":

Say goodbye to manual parameter tuningOffline, open source and free, just cue the words and pictures to let the magic happen!

Some netizens also lamented: Even a novice can make good use of the effect of the Stable Diffusion XL model.

So, how does this new image AI tool actually produce results? We tried it out.

Colab produces images in half a minute, and the effect is comparable to SD

From the operation interface, Fooocus has a total ofOver a hundred built-in stylesAvailable.

△Photo source Twitter @camenduru

As forefficiencyFooocus is also very fast in drawing. In Colab, it only takes about half a minute to draw each picture in speed priority mode:

Let's draw it first.comicsTry it and see what the "Maza War" looks like from an AI perspective. (Not this Mazaha)

Since there are still some problems with hand processing when using AI to directly generate portraits, we simply let Musk and Zuckerberg wear gloves:

It seems to work pretty well. I don't know if they have any bets, but why not let the loser dress up as a woman?

(Warm reminder: There is no winner in a fight)

In the end, the two of them "shake hands and make peace", and this precious scene was also recorded by the photographer. Does the overall picture have that flavor?

After the "Battle of Maza" ended, Ma obediently returned to the company and started selling Tesla.

neglectLOGOIfposterThe design sense is also quite online.

In fact, each of Fooocus's built-in styles is interesting, so why not take a look at these?Different stylesLive pictures:

In imitating famous works, there areCyberpunkversion,Zeldaversion,Minecraftversion, or evenPokémonVersion of Musk is available for viewing.

As for other art forms, there arePixelandLowpolystyle, andNendoroidandPaper CuttingVersion……

Of course, there are endless examples, and readers can experience more styles on their own.

(I have to say, the text in the AI-generated images is finally not just a ghost painting)

Are you all eager to try it out? We will introduce you to the gameplay of Fooocus right now!

The interface of Fooocus is like this, which can be said to be very simple:

If you just want to try it out and don't have any special requirements,This prompt box is completely sufficient..

Because the author has internalized many complex techniques into the program, operations such as parameter adjustment no longer need to be done manually.

Simply enter Prompt in the box below, click the Generate button, and wait for the image to appear.

(By default, two images are output at a time, the size is 1152×896, the style is cinematic default, and speed is prioritized)

If you need advanced settings, check Advanced in the lower left corner. The configuration information will appear on the right side of the page, divided into three tabs:

The following items can be adjusted:Size, quantity, style, performanceetc.

If you are a professional player, you can also choose the model version and even adjust the LoRA parameters.

In addition, there areAdjust sharpnessThis is high-end gameplay.

For the same content, the following animated picture shows the change of sharpness from 2 to 10 and then to 20. It can be seen that as the sharpness increases, the details of the picture become richer and richer:

However, whether Fooocus supportsChinese, we also tried it, unfortunately it doesn't work now.

For example, if we input the prompt word "apple", the resulting picture is a girl.

This... does it mean "You're the apple of my eyes"?

Now you probably know how to play Fooocus, but how to configure it?

If you have an Nvidia graphics cardWindowsIf you don't have a device, you can use the out-of-the-box version. (This is probably the 114,514th time that Lao Huang has won)

At the same time, the hardware needs to meet the minimum configuration requirements - 4GB video memory + 8GB memory.

First download directly from here:
https://github.com/lllyasviel/Fooocus/releases/download/release/Fooocus_win64_1-1-10.7z

After decompression, double-click run.bat to run it. The system will automatically download the model and deploy it. You can use it after the configuration is completed.

LinuxThe configuration requirements of this version are the same as those of Windows, but the configuration process is more complicated.

(If you have Jupyter, you can also refer to the note file used in Colab)

First, install the environment dependencies:

<span class="hljs-attr">git</span> <span class="hljs-string">clone https://github.com/lllyasviel/Fooocus.git</span>
<span class="hljs-attr">cd</span> <span class="hljs-string">Fooocus</span>
<span class="hljs-attr">conda</span> <span class="hljs-string">env create -f environment.yaml</span>
<span class="hljs-attr">conda</span> <span class="hljs-string">activate fooocus</span>
<span class="hljs-attr">pip</span> <span class="hljs-string">install -r requirements_versions.txt</span>

Then download the model file and store it in the specified directory:

△Please refer to the GitHub page for details

Of course, you can also let the system download the model automatically:

<span class="hljs-selector-tag">Python</span> <span class="hljs-selector-tag">launch</span><span class="hljs-selector-class">.py</span>

If you are usingMac, or if the hardware configuration does not meet the requirements, you can also run it directly with Colab.

(Portal: https://colab.research.google.com/github/lllyasviel/Fooocus/blob/main/colab.ipynb)

However, I still have to complain that the Colab version crashes from time to time, either automatically stopping or overflowing the memory...

If you want to run Fooocus more smoothly on a Mac or AMD computer, you can wait for the author's update.

Overall, Fooocus has good graphics output. If the prompt words are chosen well, it can even be used as Stable Diffusion. The key is that the hardware configuration requirements are not high.

How is this achieved?

Latest projects from ControlNet authors

In terms of architectural design, Fooocus is mainly divided into two parts: the interactive interface and the AI model.

in,Interactive interfaceTwo projects were referenced, namely stable-diffusion-webuiand ComfyUI.

stable-diffusion-webui is mainly the front-end design of the interactive interface:

ComfyUI combines the GUI and backend design of Stable Diffusion:

As forAI Models, we can see that the new SDXL model of Stable Diffusion is used:

This is currently one of the best versions of Stable Diffusion, and the generation effect has been greatly improved compared to the previous 1.5 version.

However, although the model and UI design of Fooocus are based on the existing Stable Diffusion open source project, the author has incorporated a lot of his own optimization designs during the production process, making the model run more smoothly.

For example, the author carefully adopted the advanced k-diffusion sampling method designed by himself in Fooocus, which can improve sampling continuity, reduce performance loss and improve sampling efficiency;

In addition, the author also carefully adjusted the parameters of the sampler and modified and added some new settings including movie style based on the original version.

The reason why the LoRA option is added to Fooocus is that the author's test found that the SDXL model with LoRA (weight set less than 0.5) almost always performs better than the SDXL model without LoRA.

The author who developed the Fooocus project is namedLvmin Zhang, graduated from Soochow University with a bachelor's degree in 2021 and is currently a doctoral student at Stanford University.

Several of his projects, including ControlNet and style2paints, have become popular: