Troubleshooting

Common issues and solutions

Installation Issues

GPU Not Detected

Symptoms: ALICE falls back to CPU, slow generation

Solutions:

  • Verify CUDA/ROCm installation
  • Check PyTorch GPU support: python -c "import torch; print(torch.cuda.is_available())"
  • Reinstall PyTorch with correct GPU support

Out of Memory Errors

Symptoms: Generation fails with OOM error

Solutions:

  • Reduce image resolution
  • Lower cache_size in config
  • Enable auto_unload
  • Close other GPU applications
  • Use smaller models (SD 1.5 instead of SDXL)

Generation Issues

Slow Generation

Causes:

  • CPU fallback (no GPU)
  • Large resolution
  • Too many steps
  • Complex model

Solutions:

  • Enable GPU acceleration
  • Use native resolution for model
  • Reduce steps (20-30 is usually enough)
  • Try Turbo models for faster results

Poor Quality Images

Solutions:

  • Increase steps (try 30-50)
  • Adjust guidance scale (7-10 recommended)
  • Use better prompts (be specific)
  • Add negative prompts
  • Try different schedulers
  • Use higher quality models

API Issues

Connection Refused

Solutions:

  • Verify ALICE is running
  • Check port (default 8080)
  • Check firewall rules
  • Verify bind address in config

Authentication Errors

Solutions:

  • Verify API key is correct
  • Check Authorization header format
  • Confirm auth is enabled in config

Next Steps