itch.io

Wow... Be sure to have updated AMD gpu drivers.
This seems to be the same problem as this : https://forum.godotengine.org/t/my-godot-is-broken-i-think/137264/16 and updating drivers worked for glebka.

Otherwise, everything looks fine so I don't know.

Nice !
I will work on the Linux version when the Hotscreen 1.0 will be ready, so I know all the features I have to adapt to Linux. I will maybe ask you some help because I have never used X11 and wayland.

What did you use to annotate images ? I manually annotated my dataset myself... xD

Unless your images are licensed under Creative Commons, I fear to have legal issues of using copyrighted images. Maybe it's stupid considering that many people are selling Loras (image generation models) trained on copyrighted images, but I will try to stay as clean as possible.

I agree that it would be nice, for example to display effects only on the feet of the woman and not the feet of the guy. However it can be difficult to determine the gender of a person with the current AI models, especially if the person is half visible, and considering that there are femboy/futa.

I guess it cost nothing to always detect a gender on all body part, it can be useful and the user would just select both genders to revert to the current operating mode.

This is exactly what I plan to do for 0.9.2 ! I hope to release the new model in 2 to 4 weeks.

Ok, I didn't get it the first time. I think I found what causes your menu window to break (not fully going invisible) when you close it.

That should be fixed in the next update.

❤️ and thank you for the suggestion.

I will add the possibility of loading your manually installed provider.

You don't have to re-install MIGraphX because it is a package managed by Windows. It is shared between apps so you don't have to re-download it every time.

It seems that AMD users must use the driver version 25.10.13.09 which can be download here : https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-25-6-... (Honestly, don't try it if you don't know what you are doing, as updating drivers can sometimes create issues.

Thank you for your feedback !

Here is my current code in the shaders that have the adapt size option :

float adapted_blur_size = blur_size;
if (adapt_size == 1) {
// estimating screen UV → polygon UV
    float uv_to_px_x = dFdx(UV.x) / SCREEN_PIXEL_SIZE.x;
    float uv_to_px_y = dFdy(UV.y) / SCREEN_PIXEL_SIZE.y;
const float typical_box_size_over_screen_size = 0.2;
    float poly_scale = typical_box_size_over_screen_size * max(uv_to_px_x, uv_to_px_y) ;
    adapted_blur_size = blur_size / poly_scale;
adapted_blur_size = clamp(adapted_blur_size, adapt_size_min_max.x, adapt_size_min_max.y);
}

because  dFdx(UV.x) is the variation of UV between two neighboring pixels so we can estimate the total size of the polygon on the screen. You can do the same in your custom fragment shader.

Note that this "works" only because polygons are only boxes for now. I do this to avoid sending the bounding box for each polygon as a uniform. I will add the uniform bounding_box_size when censor polygons could have various shape.

Thank you. I managed to fix the flickering blackboxes (reason :  my new transparency system), the recall bodypart (reason : 0.9 avoid allocating RAM so the previous frames were overridden) and inverted box on videos (reason : a change in godot 4.6).

The itch.io account that downloaded the 0.8 is allowed to download the 0.9 for free.

Thank you for confirming that it works !

You could try to delete the onnxruntime.dll file next to hotscreen.exe and then start hotscreen.

Ok I will look at that.  I guess I could also add the tensorRT dll in the zip. Thanks a lot !

The auto-install of the accelerator is only supported on Windows 11 build 26100 or newer.

I understand if you don't want to update Windows. I will look at how I can improve that.

Start the app from a command line ./hotscreen.exe

This could display more info about the crash.

The auto-install of the accelerator is only supported on Windows 11 build 26100 or newer. I will look at how I can improve that.

The auto-install of the accelerator is only supported on Windows 11 build 26100 or newer. I will look at how I can improve that.

AMD cards have the MIGraphX accelerator, which will work with Hotscreen. I just don't know how good it will be.

If you can keep the 2 gpus inside your PC you could use the RTX for the detection while the AMD card is free to run your games.

Yeah, I just saw that this provider-loader library requires Windows 11 PCs running version 24H2 (build 26100) or greater. This is why many people can't use it.

With your solution, people would still have to install/update the providers them self right ? I guess the app could give them a command line to download a provider with winget...

I will release Hotscreen 0.9 this week. It's mainly a 2x performance boost, quality of life features and bug fixes. In fact, it is ready but I will look at bugs reports that I can fix and test the build before posting it.

I wrote an explanation here : https://itch.io/t/6277931/does-anybody-know-when-the-next-update-is-coming-out

I will release Hotscreen 0.9 this week. It's mainly a 2x performance boost, quality of life features and bug fixes. In fact, it is ready but I will look at bugs reports that I can fix and test the build before posting it.

I'm sorry I haven't posted for the past two months. If you want some explanations :

•  It was already the case for previous releases : when the scheduled date is here, I feel that I didn't put enough new things (because there is so much that can be added to Hotscreen) so I'm trying to work a bit harder, but it's only delaying the release.
•  I tested many new tech/models that could be usefull for Hotscreen, so this takes time without immediately adding new features.
•  I fought with a Microsoft module.
•  I saw Hotscreen clones that seems to sell better while having less features. It not fun to compete versus vibecoding and social medias.

Ultimately, I was afraid of disappointing you. So I didn't want to look at the messages in case someone complained. But I don't deserve you guys, I'm working on this for 1 year and I never got a single angry message. You are the best. 😭

One last thing, because this post seems "sad" : I'm happy. I just pushed me a bit too hard. I hope you will still enjoy the next updates !

If you are recording your screen with OBS, once you added the recordable window as a source, you can click Hide the window  (next to create recordable window) so it will stop flashing.

Note that you can turn set the Screen latency to correct delay to 1, so the recordable window also contains your screen.

Ok 👌 , I will modify how the save timer is handled so you can call write_hotscreen_save() manually from a mod.

Try to set the Max FPS of Detection at 10 to reduce the cpu usage. You can also set the detection device to the integrated GPU (in the Detection tab).

But yeah, performances wont be great on this hardware.

If you are using a Windows x64 : Install this official Microsoft Visual C++ Redistributable https://aka.ms/vc14/vc_redist.x64.exe

Indeed, as shown in the twitter api mod, you can use this to get the running collections

for collection in HS.overlay.main_collection.collections:
    # do things to collection, note that you can also get the sub-collections with collection.collections

Use this to start/stop a collection :

collection.restart_collection()
collection.stop_collection()
collection.is_stopped()

I would have never find this without your post. I managed to solve this incorrect default behavior for the next update.

Thank you !

Starting Hotscreen from a terminal could display more infos about what causes your crash.

Using VR is tricky. The software of your VR headset must be able to get the regular Windows Desktop and add it the Recordable window (button in the Tools tab). I don't know if the Quest3 allows this kind of compositing.

The regular Hotscreen overlay have to be in "streaming app mode" so the detection can see what's under the censor.

mp3 wav and ogg should work, even if they are big. You only have to put your file anywhere in CUSTOM_DATA and check the files in the sound filter parameters. I have started to create a drag & drop system to make this easier.

In CUSTOM_DATA/Built-in mods/detection_notification.mod.tscn you have an example of the script for a mod that display a notification on your screen when a selected body part is detected. You can copy paste this to a powerful AI and ask it to modify the script so it sends a message to someone instead, through the messaging system you desire.

I will release the 0.9 in March. It's mostly huge performance boost : as games logic and graphical effects can easily be customized with AI-made mods, I focused on improving the core of Hotscreen.

You are speaking about the menu window, if I'm correct ?
(This might be because you closed the app before the menu load, so it registered a size of 0x0 pixels for the menu in the save.)

I will fix this, thank you for the report !

Yes ! I'm currently testing the newer Detection Transformer models, which claims better results than yolo models.

Have you tried the setting of filters: Stay detected during X ? I could add a "fade away" option to it.

You don't have to pay again. Just use the same itch.io account you used to buy Hotscreen the first time.

Indeed, I have the same thing if I use the Photos app from Microsoft. This seems to be a problem from Photos, which bypass the fact that the overlay should always be the topmost window. Do you experience this with other applications ?

Even using the Always on top PowerToys from Microsoft didn't solve this. 

For now, the best fix I see is to install and use the Legacy Photos from Microsoft which doesn't have this problem and mostly looks the same. XnView is also a very good image viewer.

Your ChildTest1 scene is stored in another .tscn file (ChildTest1.tscn) than RootTest.mod.tscn. This means that RootTest will try to load  res://CUSTOM_DATA/modsdev/ChildTest1.tscn (which work in Godot because the editor consider that every files of the project are inside res://) but Hotscreen can't find it with this name.

Solution 1 : if you can, all your mod should be 1 single scene. In your example, you can probably delete ChildTest1, go to the ChildTest1 scene, copy all the scene tree, and paste it under VBoxContainer in RootTest scene.

Solution 2 : you can load your other .tscn files in your script (that what my mods example add Collection and add Filter do) Like this :

var scene_for_mod = load("./CUSTOM_DATA/modsdev/ChildTest1.tscn")  # there is no res:// so Hotscreen knows its a file path
$VBoxContainer.add_child(scene_for_mod.instantiate())

It's not really a copy paste but you can save your custom filter (save button is next to its name) then click Add a filter -> Custom filter.