Stable Diffusion Not Working? Common Errors & Fixes
The 10 Stable Diffusion problems users hit most often, why they happen, and the fixes that actually work — collected from user reports, forums, and official support docs.
Stable Diffusion CUDA Out of Memory Error
Workaround availableWindows, Linux (NVIDIA GPUs)The GPU ran out of VRAM for the image size, batch or model you chose. Lowering resolution and batch size, or launching with --medvram or --lowvram, usually fixes it.
Stable Diffusion "Torch is not able to use GPU" Error
Workaround availableWindows, LinuxThe web UI installed a PyTorch build that cannot see your GPU. Rebuild the venv with the right PyTorch for your card and update drivers; --skip-torch-cuda-test only hides the check and runs on CPU.
Stable Diffusion Black Images and NaN Error
Workaround availableWindows, Linux, macOSBlack or green images and 'A tensor with all NaNs was produced' come from half-precision math failing, most often in the VAE. Launch with --no-half-vae first, then --no-half if needed.
Stable Diffusion Model Failed to Load, Exiting
Workaround availableWindows, LinuxThe checkpoint is corrupt, in the wrong folder, or the machine ran out of memory loading it. Re-download the file, put it in models/Stable-diffusion and try --lowvram.
Stable Diffusion Very Slow on Mac (M1, M2, M3)
OngoingmacOS (Apple Silicon)On Apple Silicon Stable Diffusion runs through PyTorch MPS and is much slower than on NVIDIA. Use SD 1.5 or smaller sizes, set the MPS fallback flag, or switch to a native app like Draw Things.
Stable Diffusion Not Using AMD GPU (Slow, CPU Only)
Workaround availableWindows, Linux (AMD GPUs)The original web UI targets NVIDIA CUDA. On AMD use the DirectML fork on Windows (--use-directml) or ROCm on Linux, otherwise generation runs on the CPU.
Stable Diffusion xformers Error After Update
Workaround availableWindows, Linuxxformers must match your exact torch and CUDA versions. After an update, reinstall matching versions or run without xformers (the 'No module xformers' notice is harmless).
Stable Diffusion "Couldn't install torch" Error
Workaround availableWindows, LinuxThe first launch failed to download or install PyTorch. Check the Python version, delete the venv, and rerun, or use Forge's one-click package.
Stability AI API 402 Insufficient Credits Error
OngoingStability AI APIA 402 means the account has run out of credits. Add credits in the platform account; retrying does not help.
Stability AI API 429 Too Many Requests
Workaround availableStability AI APIStability's API allows 150 requests per 10 seconds; going over returns 429 and a 60-second timeout. Throttle and back off.
Problem not listed? See all Stable Diffusion complaints or compare Stable Diffusion alternatives.
// We build software. Any software.
Have an idea for an app? We'll build it.
We're a small team of developers who build custom software fast. Not just tools like Stable Diffusion: web and mobile apps, games, desktop software, AI tools, internal tools: whatever you have in mind. 50+ custom products built for clients so far, and we run our own products too: booking platforms, games, learning apps. See our work
- Web apps & SaaS
- Mobile apps (iOS & Android)
- Games
- Desktop software
- AI tools & automations
- Internal tools & dashboards
- Client portals & booking
- Your own version of any tool
- ✓ Free 15-min call to scope it
- ✓ Price agreed up front
- ✓ Your data, your code