[Fixed] RetroDiffusion does not work for RTX 5000/6000 GPUs

As of 5/31/2025, the RetroDiffusion extension for Aseprite does not work out of the box if you have an Nvidia RTX 5000 or RTX PRO 6000 series graphics card.

The issue is that RetroDiffusion creates a virtual environment and installs an outdated version of PyTorch that is not compatible with these newer GPUs.

Specifically, it installs:

safetensors==0.4.2
torch==2.3.0 -i https://download.pytorch.org/whl/cu121 --trusted-host download.pytorch.org
torchvision==0.18.0 -i https://download.pytorch.org/whl/cu121 --trusted-host download.pytorch.org

Here, cu121 refers to CUDA version 12.1, which does not support the RTX 5000 and RTX PRO 6000 series GPUs. You can check GPU compatibility on Nvidia's CUDA GPU Compute Capability page. If you try to run Retro Diffusion in Aseprite, you would get following error.

RuntimeError: CUDA error: no kernel image is available for execution on the device

To fix this, you'll first follow the normal RetroDiffusion setup process, then update the PyTorch/CUDA version via the terminal.

Steps

Make sure you have completed the initial setup for the RetroDiffusion extension.

1. Quit Aseprite.

2. Open Terminal.

3. Navigate to the stable-diffusion-aseprite folder.

    cd ~\AppData\Roaming\Aseprite\extensions\RetroDiffusion\stable-diffusion-aseprite
   

4. Check your CUDA version.

    nvidia-smi
   

   You should see something like CUDA Version: 12.8. Take note of this version. You'll use it in Step 7.

5. Activate the virtual environment.

    .\venv\Scripts\Activate.ps1
   

6. Check the installed version of PyTorch.

    pip show torch
   

   It will likely show something like Version: 2.3.0+cu121.

7. Go to the official PyTorch install page and select your correct configuration (e.g., pip, Python, CUDA 12.8).

   

8. Copy and paste the pip3 install command from the website into your terminal. For CUDA 12.8, it should look like:

    pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128
   

9. Verify the update.

    pip show torch
   

   It should now show something like Version: 2.7.0+cu128.

10. Launch Aseprite and run RetroDiffusion to generate images. Done. Enjoy!