Linux fixes and CUDA support for WaveSpeed Desktop

[tchung1970]

Hello,

I've been using WaveSpeed Desktop on Ubuntu 24.04 and encountered a few issues that prevent it from working out of the box. I've documented the problems and solutions, and built CUDA binaries that work with my NVIDIA GPU (RTX 4000 Ada Generation).

Issues found:

1. App crashes on launch due to Electron sandbox incompatibility with Ubuntu 24.04
2. SD binary fails to find libstable-diffusion.so (Linux library path issue)
3. Pre-built AVX512 binary causes "Illegal instruction" on many CPUs

I've created a repository with detailed fixes and pre-built CUDA binaries:
https://github.com/tchung1970/wavespeed-desktop-cuda

The DEVELOPER-GUIDE.md includes exact code changes for each fix, CUDA build instructions, and a GitHub Actions workflow example.

Would you please consider incorporating these fixes and adding CUDA support to the official Linux release?

Thanks for building this great app!

ps. You can contact me at tchung1970@gmail.com

[chengzeyi]

@tchung1970 Thanks! We will take a look into this. We have tried many times to fix the sandbox problems but looks like still failed in your cases. Also, we could prefer Vulkan over CUDA since it is more general (also works with non-NVIDIA cards). Do you have any suggestions?

[chengzeyi]

@codex Create a PR to solve the crash issue and linux library path problem at first

Take a look at https://github.com/tchung1970/wavespeed-desktop-cuda/blob/main/DEVELOPER-GUIDE.md

[chatgpt-codex-connector]

Summary

• Replaced the Linux post-install symlink with a wrapper that forces --no-sandbox, fixes the desktop entry, and refreshes icons.
• Limited Linux packaging to the deb target to align with the new wrapper-based launch flow.
• Set LD_LIBRARY_PATH when spawning the SD binary so bundled shared libraries are discovered.

Testing

• ⚠️ Not run (not requested).

View task →

[tchung1970]

Hello,

How about this approach? I created a modular CUDA plugin for wavespeed-desktop
https://github.com/tchung1970/sd-cli-cuda

Approach:

• Standalone .deb package - Users install separately from wavespeed-desktop
• Auto-integration - Postinst script creates symlink at ~/.config/wavespeed-desktop/sd-bin/sd pointing to /usr/bin/sd-cli
• No changes needed to wavespeed-desktop - It already checks that path for the backend

Benefits:

• Users with NVIDIA GPUs can easily add CUDA acceleration
• Keeps wavespeed-desktop lightweight (no bundled binaries)
• Easy to update sd-cli-cuda independently
• Clean uninstall via dpkg -r sd-cli-cuda

Built with:

• stable-diffusion.cpp compiled with CUDA 12.0
• Tested on RTX 4000 Ada Generation

Would you mind mentioning my name in your next release notes for credit? :)
Thank you.

[tchung1970]

Hello,
I've confirmed version 1.0.40 works great with my sd-cli-cuda from https://github.com/tchung1970/sd-cli-cuda
Thank you.

Generation failed

$ wavespeed-desktop
08:32:22.539 › ================================================================================
08:32:22.545 › Application starting...
08:32:22.545 › Version: 1.0.40
08:32:22.545 › Platform: linux x64
08:32:22.546 › Electron: 33.4.11
08:32:22.546 › Chrome: 130.0.6723.191
08:32:22.547 › Node: 20.18.3
08:32:22.547 › Log file: /home/tchung/.config/wavespeed-desktop/logs/main.log
08:32:22.547 › ================================================================================
08:32:22.549 › Checking for beta autoupdate feature for deb/rpm distributions
08:32:22.549 › Found package-type: deb
08:32:22.717 › Loading renderer from: /opt/WaveSpeed Desktop/resources/app.asar/out/renderer/index.html
08:32:22.718 › File exists: true
08:32:25.730 › Checking for update
08:32:25.747 › Generated new staging user ID: 9094f91c-5d25-5779-aaf1-21d428322bba
08:32:26.897 › Update for version 1.0.40 is not available (latest version: 1.0.40, downgrade is allowed).
08:32:41.015 › [System Info] Platform: linux, Acceleration: CUDA
08:32:51.478 › [SD Extract] Extracting: /home/tchung/.config/wavespeed-desktop/sd-bin/sd.zip
08:32:51.479 › [SD Extract] Destination: /home/tchung/.config/wavespeed-desktop/sd-bin/sd
08:32:51.814 › [SD Extract] Extracted to temp directory: /home/tchung/.config/wavespeed-desktop/sd-bin/temp-extract
08:32:51.815 › [SD Extract] Found binary: /home/tchung/.config/wavespeed-desktop/sd-bin/temp-extract/sd-cli
08:32:51.815 › [SD Extract] Binary directory: /home/tchung/.config/wavespeed-desktop/sd-bin/temp-extract
08:32:51.815 › [SD Extract] Copying all files to: /home/tchung/.config/wavespeed-desktop/sd-bin
08:32:51.816 › [SD Extract] Copied: ggml.txt
08:32:51.870 › [SD Extract] Copied: libstable-diffusion.so
08:32:51.871 › [SD Extract] Copied: sd-cli
08:32:51.873 › [SD Extract] Copied: sd-server
08:32:51.873 › [SD Extract] Copied: stable-diffusion.cpp.txt
08:32:51.874 › [SD Extract] Created sd copy for compatibility
08:32:51.934 › [SD Extract] Made files executable
08:32:51.942 › [SD Extract] Cleanup completed
08:33:59.518 › [SD Generate] Using downloaded binary: /home/tchung/.config/wavespeed-desktop/sd-bin/sd
08:33:59.518 › [SD Generate] GPU VRAM detection unavailable
08:33:59.520 › [SDGenerator] Spawning SD process: /home/tchung/.config/wavespeed-desktop/sd-bin/sd
08:33:59.520 › [SDGenerator] Arguments: ["--diffusion-model","/home/tchung/.config/wavespeed-desktop/models/stable-diffusion/z_image_turbo-Q4_K.gguf","-p","a fancy letter Z","-W","1024","-H","1024","--steps","4","--cfg-scale","1","-o","/home/tchung/Documents/WaveSpeed/images/local_z-image_mjpy8nh75dev_0.png","-v","--offload-to-cpu","--diffusion-fa","--llm","/home/tchung/.config/wavespeed-desktop/models/stable-diffusion/auxiliary/Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf","--vae","/home/tchung/.config/wavespeed-desktop/models/stable-diffusion/auxiliary/ae.safetensors","-n","blurry, bad quality, low resolution, watermark, distorted, ugly, deformed, extra limbs, poorly drawn, bad anatomy","--seed","2044533257","--sampling-method","euler","--scheduler","simple"]
08:34:03.924 › [SDGenerator] Generation failed: Process exited with code unknown

Before sd-cli-cuda

$ ls -l ~/.config/wavespeed-desktop/sd-bin/
total 26220
-rwxrwxrwx 1 tchung tchung 1078 Dec 28 08:32 ggml.txt
-rwxrwxrwx 1 tchung tchung 24371816 Dec 28 08:32 libstable-diffusion.so
-rwxrwxrwx 1 tchung tchung 664072 Dec 28 08:32 sd
-rwxrwxrwx 1 tchung tchung 664072 Dec 28 08:32 sd-cli
-rwxrwxrwx 1 tchung tchung 1128384 Dec 28 08:32 sd-server
-rwxrwxrwx 1 tchung tchung 1062 Dec 28 08:32 stable-diffusion.cpp.txt

Install sd-cli-cuda

$ sudo dpkg -i sd-cli-cuda_1.0.0_amd64.deb
Selecting previously unselected package sd-cli-cuda.
(Reading database ... 200404 files and directories currently installed.)
Preparing to unpack sd-cli-cuda_1.0.0_amd64.deb ...
Unpacking sd-cli-cuda (1.0.0) ...
Setting up sd-cli-cuda (1.0.0) ...
Created symlink for user: tchung

sd-cli-cuda installed successfully!
Binary: /usr/bin/sd-cli
Symlinks created in ~/.config/wavespeed-desktop/sd-bin/sd

After sd-cli-cuda

$ ls -l ~/.config/wavespeed-desktop/sd-bin/
total 25568
-rwxrwxrwx 1 tchung tchung 1078 Dec 28 08:32 ggml.txt
-rwxrwxrwx 1 tchung tchung 24371816 Dec 28 08:32 libstable-diffusion.so
lrwxrwxrwx 1 tchung tchung 15 Dec 28 08:57 sd -> /usr/bin/sd-cli
-rwxrwxrwx 1 tchung tchung 664072 Dec 28 08:32 sd-cli
-rwxrwxrwx 1 tchung tchung 1128384 Dec 28 08:32 sd-server
-rwxrwxrwx 1 tchung tchung 1062 Dec 28 08:32 stable-diffusion.cpp.txt

Generation Successful

$ wavespeed-desktop
08:58:20.645 › ================================================================================
08:58:20.651 › Application starting...
08:58:20.651 › Version: 1.0.40
08:58:20.652 › Platform: linux x64
08:58:20.652 › Electron: 33.4.11
08:58:20.652 › Chrome: 130.0.6723.191
08:58:20.653 › Node: 20.18.3
08:58:20.653 › Log file: /home/tchung/.config/wavespeed-desktop/logs/main.log
08:58:20.653 › ================================================================================
08:58:20.655 › Checking for beta autoupdate feature for deb/rpm distributions
08:58:20.656 › Found package-type: deb
08:58:20.827 › Loading renderer from: /opt/WaveSpeed Desktop/resources/app.asar/out/renderer/index.html
08:58:20.827 › File exists: true
08:58:23.841 › Checking for update
08:58:24.952 › Update for version 1.0.40 is not available (latest version: 1.0.40, downgrade is allowed).
08:58:27.425 › [System Info] Platform: linux, Acceleration: CUDA
08:58:27.469 › [SD] Using downloaded binary: /home/tchung/.config/wavespeed-desktop/sd-bin/sd
08:58:34.506 › [SD Generate] Using downloaded binary: /home/tchung/.config/wavespeed-desktop/sd-bin/sd
08:58:34.506 › [SD Generate] GPU VRAM detection unavailable
08:58:34.508 › [SDGenerator] Spawning SD process: /home/tchung/.config/wavespeed-desktop/sd-bin/sd
08:58:34.508 › [SDGenerator] Arguments: ["--diffusion-model","/home/tchung/.config/wavespeed-desktop/models/stable-diffusion/z_image_turbo-Q4_K.gguf","-p","a fancy letter Z","-W","1024","-H","1024","--steps","4","--cfg-scale","1","-o","/home/tchung/Documents/WaveSpeed/images/local_z-image_mjpz49l2ns2h_0.png","-v","--offload-to-cpu","--diffusion-fa","--llm","/home/tchung/.config/wavespeed-desktop/models/stable-diffusion/auxiliary/Qwen3-4B-Instruct-2507-UD-Q4_K_XL.gguf","--vae","/home/tchung/.config/wavespeed-desktop/models/stable-diffusion/auxiliary/ae.safetensors","-n","blurry, bad quality, low resolution, watermark, distorted, ugly, deformed, extra limbs, poorly drawn, bad anatomy","--seed","1940089893","--sampling-method","euler","--scheduler","simple"]
08:58:48.062 › [SDGenerator] Generation successful