Update 0.9
Big performance update
In short : same quality, 2x faster, 2x less screen-censor latency, 2x less cpu/gpu usage.
Faster detection thanks to the GPU-specific accelerator
AI models can run faster on your GPU thanks to certain specific software (e.g., tensorRT for Nvidia GPUs). Hotscreen allows you to install this accelerator in one click. This is a lightweight Windows module from Microsoft. Detection is twice as fast on my Nvidia RTX graphics card. I don't know if this will be the case for AMD and Intel cards, but the point is that performance should improve as manufacturers update their accelerators.
Select the accelerator corresponding to your GPU and click Install. The accelerator will now be available. Select it. Done ! (The first time TensorRT loads a model can take 5 sec to optimize it for your GPU but it will be instant the following times.)
CPU usage has been halved for the Windows 10/11 capture method. This also results in a halved latency and a double sampling rate up to 300Hz.
Memory management could cause some freezes, especially with large screen. This was fixed : CPU usage is reduce and system RAM usage won't go above 1 GB.
Detect small bodyparts x4 option is now faster. Preprocessing of screen images for detection now uses threads to speed up calculations, especially for 640x640 models.
All of this reduces the Hotscreen's impact on your PC and lowers latency. It also allows the use of more precise detection models, such as 640x640 and the upcoming powerful models I plan to release.
Quality of life
The screen latency will never automatically reset to 0. It will only be temporarily disabled in case the screen recording is too slow.
You can now drag & drop files on images/audios/videos filters. Non .ogv videos will even be automatically converted if ffmpeg is installed.
Drag & drop the files on Image/Video/Sound filter to auto-import them. This will create copies inside CUSTOM_DATA.
New parameters
You can configure the boxes so that they gradually disappear when their detection is lost.
Fast text filter can now disable the randomized position of the text.
You can now use {...} in levels name to perform operations. For example, the level name Next level in {100 - progress} would display a countdown.
Bugfixes
Fixed a bug where manual boxes for one filter where also used in other filters.
Fixed a bug where taking a screenshot could disable the screen latency or screen effects.
Fixed a bug where the menu window would not appear if the menu had been previously closed on another screen.
Fixed a bug where the DPI scale of Hotscreen was incorrect when using monitors with different scale values.
Modding : calling HS.config_reader.write_hotscreen_save() won't impact anymore the frequency of auto-saves.
Modding : Hotscreen now uses Godot 4.6.3, so you can create mods with any Godot 4 version up to this one.
Comments
Log in with itch.io to leave a comment.
Does anyone else feel it's slightly worse at detecting hentai/furry? Other than that, the improvements feel good.
This is by far the best censorship software. Thank you!!! Been trying this new update on my 3050 laptop and it runs buttery smooth, keep up the good work :)!
Thank you for the Update!!
I'm about to switch from NVIDIA RTX 2060 to AMD 9070 XT. The only thing that concerns me is the CUDA being used in lots of tools like those tensorRT or other software based on trained models, and since the 0.9 update, i wonder if there's any alternative of tensorRT RTX for AMD cards available? I know that even on the raw rasterization power after the upgrade im gonna achieve higher detection framerates, but im aware that nvidia in that topic is far more superior and im just curious and eager to ask if i'm gonna be able to play new video game titles on ultra and have the hotscreen running nicely with it. :D
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.
I've tried many similar apps and Hotscreen is without a doubt the best, keep up the good work man
I didn't have an option to select model and it prompted me to install windows app runtime. After installing it and restarting app I get:
"Error while listing providers. Please check hotscreen_logs.txt"
This is logs file:
Hotscreen using profile : default
Hotscreen Overlay starting at Max FPS : 180
Translations added from ./CUSTOM_DATA/translations.csv — languages : ["en", "zh", "es", "fr", "ar", "ru", "pt", "ko", "ja", "de"]
Using saved language code : pl
Windows App SDK successfully initated : 8000.859.21.0
Onnx Runtime version 1.24.3
Error : no WindowsML providers are founds
Windows version : Microsoft Windows NT 10.0.22631.0
CPU : Intel(R) Core(TM) i7-14700K, Cores : 20, Threads : 28, Max clock : 3100 MHz
Using video adapter : NVIDIA GeForce RTX 5080 from NVIDIA type DiscreteGpu
Driver infos : d3d12 version 12_0 api 12_0
Adapter 0 NVIDIA GeForce RTX 5080
Screen Rect 0 : Rect { X = 0, Y = 0, Width = 3440, Height = 1440, Top = 0, Bottom = 1440, Left = 0, Right = 3440, Location = Point { X = 0, Y = 0 }, Size = Size { Width = 3440, Height = 1440 }, TopLeft = Point { X = 0, Y = 0 }, BottomRight = Point { X = 3440, Y = 1440 } } Orientation : 1
Screen Rect 1 : Rect { X = 3440, Y = 0, Width = 1920, Height = 1080, Top = 0, Bottom = 1080, Left = 3440, Right = 5360, Location = Point { X = 3440, Y = 0 }, Size = Size { Width = 1920, Height = 1080 }, TopLeft = Point { X = 3440, Y = 0 }, BottomRight = Point { X = 5360, Y = 1080 } } Orientation : 1
Adapter 1 Intel(R) UHD Graphics 770
Adapter 2 NVIDIA GeForce RTX 5080
Adapter 3 Microsoft Basic Render Driver
Starting Screen Recorder
Initializing Windows10Capture using Video card 0 and monitor 0
Hotscreen version : Windows - Full version - v0.9
Max FPS of Overlay set to 180
Windows10Capture initiated on : 3440x1440
Frame Latency of screen set to 5
Max FPS Record start at 180
Loading mod scene : ./Mods/HotscreenOverlayAlwaysOnTop-main/HsOverlayAlwaysOnTop.mod.tscn
MaxFPSRecord at 180
Recreating d3dDevice for fast WGC copy
Recreating stagingTexture for fast WGC copy
Starting DirectML session for ./models/hs-real-anime-y11n-640-fp16.onnx on device 0
Starting DirectML session for ./models/eyes-detection-01.onnx on device 0
Avx is supported.
Max FPS Detection start at 60
Use double Detection start at True
Update time of eyes detection start at 0,05
Setting MaxFPSDetection at 60
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.
Wow the performance is really good, even on a widescreen. Nicely done!