❓ FAQ & Troubleshooting

If FXTD Radiance nodes don't appear in the node menu:

• Restart ComfyUI completely — Not just refresh, but fully restart
• Check the console for error messages during startup
• Verify the folder is in custom_nodes/FXTDStudio_Radiance/
• Make sure __init__.py exists in the folder

[FXTDStudio Radiance] Loaded 50+ nodes (v1.0.0)

If you see "ModuleNotFoundError" in the console:

cd custom_nodes/FXTDStudio_Radiance
pip install -r requirements.txt

Common missing modules:

• colour-science — For advanced color transforms
• opencv-python — For image I/O including EXR
• scipy — For signal processing functions

This is just a warning, not an error. Most nodes will work fine without it.

colour-science is only needed for:

• Advanced ACES transforms
• Precise chromatic adaptation
• spectral color science operations

pip install colour-science

If nodes are running slowly or falling back to CPU:

• Ensure CUDA is installed and PyTorch can see your GPU
• Check that use_gpu is enabled on nodes that have it
• Your GPU memory might be full from other operations

import torch
print(f"CUDA available: {torch.cuda.is_available()}")
print(f"Device: {torch.cuda.get_device_name(0)}")

If you're getting CUDA out of memory errors:

• Reduce image resolution during processing
• Enable tile processing on upscale nodes
• Process one image at a time (batch size 1)
• Close other GPU-intensive applications

Some operations are inherently CPU-bound:

• EXR file saving (disk I/O)
• LUT loading (file parsing)
• First run (compilation/warmup)

Expected performance: Most GPU operations should be 20-50x faster than CPU.

This is likely log-encoded footage!

If you've used LogC encoding, the image is supposed to look flat. This preserves maximum dynamic range for color grading in DaVinci Resolve or other NLEs.

Standard 8-bit images clip at 1.0. For HDR workflows:

• Use Image to Float32 early in your pipeline
• Enable HDR Expand Dynamic Range for more headroom
• Export as 16-bit or 32-bit EXR

For more realistic grain:

• Lower the grain_intensity — less is more
• Increase grain_softness for a more organic look
• Match your subject: use lower grain for clean subjects, higher for gritty scenes
• Combine with subtle halation for authenticity

grain_intensity: 0.15
grain_size: 0.8
grain_softness: 0.4
halation: 0.1

EXR export requires OpenCV with EXR support:

pip uninstall opencv-python
pip install opencv-python-headless

The save path must also be writable. Check the console for the actual output path.

Color space mismatch! EXR files store linear data by default.

• If exporting sRGB, the receiving app may expect linear
• If exporting log, tell the receiving app which log curve

Banding in gradients is usually caused by:

• Viewing on an 8-bit monitor (can't display 16-bit)
• The source not having enough tonal range
• Aggressive color grading amplifying limited data

Common causes and fixes:

• Too few steps: Use at least 20 for flux-dev, 4-8 for flux-schnell
• Wrong scheduler: Use "simple" for Flux models
• CFG too high: Keep at 1.0 for Flux (it uses guidance differently)

steps: 20 (dev) or 4 (schnell)
cfg: 1.0
sampler: euler
scheduler: simple
flux_shift: 1.0
flux_guidance: 3.5

flux_shift adjusts the sigma (noise) schedule for Flux models.

• 1.0 — Default, good for most uses
• 1.5-2.0 — Better for higher resolutions
• 3.0+ — Experimental, can improve detail at very high res

This is similar to how SDXL uses different schedules for 1024px vs 512px.

The FXTD Radiance Sampler adds:

• Flux-specific sigma shifting — Built into the node
• Flux guidance parameter — Controls the Flux-specific CFG
• Timing diagnostics — See how long sampling takes
• Better defaults — Pre-configured for Flux

It's fully compatible with non-Flux models too!

The Radiance Pro Viewer requires JavaScript extensions:

• Make sure js/ folder exists in the extension
• Hard refresh your browser (Ctrl+Shift+R)
• Check browser console for JavaScript errors

[FXTDStudio Radiance] Radiance Pro Viewer JavaScript extension enabled

Make sure the viewer has focus:

• Click on the image in the viewer first
• No other input fields should be focused

H — Toggle histogram
W — Toggle waveform
V — Toggle vectorscope
E — Toggle false color (exposure)
F — Toggle fullscreen
R / G / B — Isolate channels
1 — Reset to 100% zoom
0 — Fit to screen

Connect two images to compare them:

• image — Primary image (A)
• compare_image — Secondary image (B)