florr auto afk (v1.3.2) (2025-07-05 Update)

As m28 released his new anti afk, I‘ll put my new anti-anti afk code here.

Web Page: https://shiny-ladybug.github.io

For developing extensions, go check DEVELOPMENT.MD

Caution

M28 can send player messages and AFK Checks. If you can solve the check but cannot respond to his messages, it may result in a BAN.

How to Use

Executables

A cat told me that somehow he can't understand this README document. So I decided to fix it. I figured I should explain how to USE it.

Go to RELEASE and click the blue stuff that says florr-auto-afk-v1.*.*.zip.

Then you can unzip it, put it anywhere you want, and then run the segment.exe file.

Install ./extension.js in Tampermonkey to enable Extensions.

Deploy

If you want the source code as well as the exe file, you might want to choose this option.

Firstly, go to whatever location you want this project to be and run cmd in it. Then run the command below to clone this repository:

git clone https://github.com/Shiny-Ladybug/florr-auto-afk

After successfully cloning this repository, run the code below to install the necessary dependencies:

pip install -r ./py311-requirements.txt

Now you can run the code:

python segment.py

For the latest v1.2.x or v1.3.x, you need WINDOWS device to run.

Also, the release version is only for WINDOWS and CPU ONLY users.

If you want to run the script on MacOS or Linux, then sadly, you can only use v1.1.1.

By the way, you can build your own GPU version here by installing torch-gpu.

Extensions

Internal

Using websocket and Tampermonkey to get and modify florr.io canvas methods.

Make sure you turn Developer mode on in Browser Settings.

Install ./extension.js into the browser and turn something suspicious on in ./config.json.

External

These extensions are stored under ./extensions

Go check DEVELOPMENT.MD for further information

LLM Configs

This will allow LLM to control your movements.

Default conversations are stored under ./conversations.json, edit them if you like

All available actions are chat, come up (come here), move (up, right, down, left), switch petal, join squad and follow

You should modify the aiChat settings manually to enable them.

Using Ollama will require to install a local Ollama, visit https://ollama.com/download for further information

ollama pull qwen3:14b
ollama run qwen3:14b

Using OpenAI will require an openai-api-compatible endpoint to request, normally the endpoint will be like v<Num>/completions

Using Random will not support aiChat.fullControl, they will just generate random responses.

Config

See the settings page in the GUI menu. In the config, you can tweak parameters to what you like.

If you are Chinese by accident and can't understand what this is, set language to zh-cn for Chinese settings.

Some cats still weren't satisfied with this and want explicit explanations for how everything in the setting works.

Now you can go to Settings.md to learn about all that stuff.

Oh, and by the way, people didn't even notice that hovering above the config choices gives more details.

INFO

In the latest v1.2+.x version, I added background AFK Check detection support. This requires that your browser supports disabling CalculateNativeWinOcclusion.

Details are below:

Chrome

Go to chrome://flags/ and search for CalculateNativeWinOcclusion. If this appears, go disable it.

If you can't find the element, try running this and replace the chrome.exe path if needed.

"C:\Program Files\Google\Chrome\Application\chrome.exe" --disable-features=CalculateNativeWinOcclusion

Edge (Not Supported)

I didn't find a method to disable the feature, however, I found this in Edge Beta, Edge Dev, and Edge Canary, check the Insider Edge version at https://www.microsoft.com/en-us/edge/download/insider

Hence, you should install one of Edge Beta, Edge Dev, or Edge Canary.

After you install the Insider version, go to edge://flags/ and disable CalculateNativeWinOcclusion (essentially the same as what you do for Chrome).

I find there is no sight of disabling CalculateNativeWinOcclusion on 05/27/2025. Both Edge Beta (137.0.3296.39) and Edge Canary (137.0.3283.0) cannot disable the feature any longer.

So I announce that Edge is End of Support now.

Firefox

Congratulations, Firefox disables CalculateNativeWinOcclusion by default, enjoy your game in Firefox.

Note

I have opened-sourced the code here.

If you want to use our models to write more automation code, go check the models in assets

If you want to boot this without an Internet Connection, try to set skipUpdate in config.json to true.

P.S. Sidenote: After trying the ONNX models, I think it's the worst decision I have ever made.

Now those stupid codes go under the onnx branch.

The Exposure

As that exposure can nerf the mobs' movement effect and idle detection enables me to active (maybe kill supers) without quitting the program

Gallery

Old model was trained on 4/30/2025 based on 107 instances with 698 epochs.

Old model was trained on 5/23/2025 based on 225 instances (192 train, 33 val) with 431 epochs.

Thanks to you guys for uploading your datasets. I soon get bunches of raw datasets in https://github.com/Shiny-Ladybug/florr-afk.

Also thanks to BF52[bot] (He was installed in the repo), he removes lots of invalid datasets for me automatically.

New model is trained on 6/01/2025 based on 878 instances (747 train actually 747 should be a boeing jet instead of a train, 131 val) with 521 epochs.

· · ·

Difficulty

I do rank those AFK Checks so that I can do sorts easily

$$ D = \sqrt{ \max\left(1, 3 * P_{density}\right) \cdot \left( \frac{P_{l}}{P_{w}} \right) \cdot \sqrt{N_{p}} } $$

The density using KDTree to count the neighbor points

How it works

1. Use afk-det.pt model to detect the AFK Check windows.

   The model contains 3 classes: AFK_Window, Start, End

   Model most recent updated on 4/30/2025

   Training for 632 epochs based on 107 instances.

2. Try to use the YOLO model afk-seg.pt to separate the AFK Path.

   Detailed Model Information go check Model Gallery.

3. Use modified dijkstra algorithm to find the path.

   If no path was found, use the cv2.ximgproc.thinning() method to get the skeletonized path.

4. Loop

FAQ

Q: Why can't I clone or download this repository?

A: Man, what can I say? Skill issue.

Q: Why can't the GUI load properly?

A: You are probably too anxious. Be patient. It takes 10s to initialize on my computer and might take several minutes on your sloth-speed computer.

Q: What is the accuracy of this model?

A: According to statistics, after properly installing the newest version of Auto AFK, the accuracy should be greater than 90%.

Q: Why is the accuracy tested on my computer so low?

A: Most likely skill issue. Make sure you are NOT using normal Edge (mentioned above). Also make sure you have disabled CalculateNativeWinOcclusion if you're on Google Chrome or other versions of Edge.

Q: Why is my computer so hot when running the code?

A: Make sure you close all the unnecessary programs when running and maybe turn on "Best Battery Life".

Q: But my computer goes to sleep after a very short period. How to fix that?

A: Go to Settings in your computer and change the sleep after inactive time, of course.

Q: I found some bugs/I have some ideas for improving this application. What should I do?

A: Suggest turning on Save Trainable Dataset in Settings > Advanced. Then contact Shiny Ladybug via QQ.

Issues

I DO NOT RECOMMEND TRUSTING THIS SCRIPT

It can pass some AFK Checks.

The longer you stay in the same server, the AFK checks get harder.

The script cannot solve the WORM-LIKE disgusting checks for the time being.

?

Some people queried why I was using GPLv3 without opensourcing.

BRO, that night was TRADITIONAL CHINESE NEW YEAR'S EVE, can't I just upload the release code and have a break at this good time. Or you can read the release notes carefully, word by word, and you can find the problem.