HAL-Net
Analog Film Halation Estimation and Synthesis
A CNN trained on 100 authentic CineStill 800T film scans. Upload any image — HAL-Net estimates halation parameters from highlight regions and synthesizes a warm analog film glow.
Inspired by FGA-NN (Ameur et al., 2025).
Manual Override
Adjust sliders to override HAL-Net predictions.
What is halation? Halation is the red-orange glow around bright light sources in analog film photography. It occurs when light passes through the emulsion, reflects off the film base, and exposes the silver halide crystals from behind. CineStill 800T is particularly known for this effect because its anti-halation layer was removed during processing for cinema use.
Evaluation (15-image test split)
| Method | Radius MAE | Intensity MAE | Warmth MAE |
|---|---|---|---|
| Mean-pred baseline | 0.012 | 0.317 | 0.328 |
| HAL-Net | 0.015 | 0.247 | 0.218 |
HAL-Net beats the baseline on the two harder parameters (intensity and warmth). Radius is already near-zero for both methods because the pseudo-label distribution is narrow — nearly every image maps to a spread around 0.09.
Dataset 100 handpicked CineStill 800T scans from Flickr and Lomography exhibiting authentic halation. Train / val / test split: 70 / 15 / 15.