📷 Camera Forensic Realism Engine v4

Makes AI-generated images look like they were shot on an iPhone 17 by emulating Apple's ISP color pipeline — with honest, physically-grounded color science.

No placebo sliders. Every stage in v4 does real, verifiable work: white balance and tone mapping in linear light, local HDR in log-luminance, color rendering in Oklab, and tetrahedral LUT interpolation.

---

✨ Processing Stages

| # | Stage | What It Does | |---|-------|-------------| | 1 | White Balance | Channel gains in linear light, luminance-preserving (warming doesn't brighten) | | 2 | Global Tone | Exposure, shadow lift, midtone contrast around the 0.18 pivot, Reinhard highlight rolloff | | 3 | Smart HDR | Durand-style local tone mapping in log-luminance — shadows up, highlights down, detail untouched | | 4 | Color Science | Oklab rendering: hue-preserving vibrance, real skin protection, iPhone split-tone (cool shadows / warm highlights) | | 5 | Detail | Single two-scale stage (texture + clarity) with tanh halo suppression, luminance-only | | 6 | Optics & Sensor | Natural 1/(1+kr²)² vignette in linear light + photon-weighted grain |

Plus 14 iPhone 17 Photographic Styles presets (undertones + moods) applied as offsets on top of your sliders.

What changed from v2/v3

• ❌ Removed the fake "Display P3" stage — it converted sRGB→XYZ→P3 and back (a mathematical no-op) and the only real effect was a plain saturation multiply. Replaced with genuine Oklab vibrance.
• ❌ Removed redundant multiplier sliders (wb_strength, tone_strength, sensor_strength) — master_strength is now the one honest mix control.
• ❌ Merged three overlapping sharpeners (detail boost / Deep Fusion / ISP sharpening) into one coherent detail stage.
• 🐛 Fixed the box blur used by HDR and sharpening — it was a trailing window that shifted every blur diagonally and darkened the top/left borders.
• 🐛 Fixed the non-monotonic highlight clip that could cause banding above white.
• ✅ Skin handling is now an actual protection mask (Oklab hue ≈ 50°, plausible chroma/lightness) instead of a brightness filter on every warm pixel.

---

🚀 Installation

Option 1: Git Clone (Recommended)

cd ComfyUI/custom_nodes
git clone https://github.com/MONKEYFOREVER2/ComfyUI-CameraForensicRealism.git
cd ComfyUI-CameraForensicRealism
pip install -r requirements.txt

Restart ComfyUI.

Option 2: Manual Download

1. Click the green Code button → Download ZIP
2. Extract into ComfyUI/custom_nodes/ as ComfyUI-CameraForensicRealism
3. pip install -r requirements.txt
4. Restart ComfyUI

---

🎯 Finding the Node

Right-click → Add Node → image/forensic → 📷 Camera Forensic Realism (iPhone 17)

⚡ Quick Start

1. Place the node after your detailers/post-processing and before SaveImage
2. Pick a Photographic Style (start with Standard)
3. Defaults are tuned for the iPhone 17 look — adjust to taste

[Generate] → [Upscale] → [FaceDetailer] → [📷 Camera Forensic Realism] → [SaveImage]

---

🔧 Parameters

Master

| Parameter | Default | Description | |-----------|---------|-------------| | photographic_style | Standard | iPhone 17 Photographic Style preset (see below) | | master_strength | 0.85 | Final blend between original and processed image | | seed | 0 | Grain reproducibility |

Photographic Styles

Moods: Standard, Natural, Luminous, Vibrant, Dramatic, Quiet, Cozy, Ethereal, Muted B&W, Stark B&W Undertones: Amber, Gold, Rose Gold, Cool Rose, Neutral

Styles are applied as offsets on top of your slider values, mirroring how Apple's 2nd-gen Photographic Styles re-tune the same pipeline.

White Balance

| Parameter | Default | Description | |-----------|---------|-------------| | enable_white_balance | True | Linear-light channel gains, luminance-preserving | | wb_temperature | 0.12 | Cool (blue) ← → Warm (orange). iPhone AWB ≈ 0.1–0.15 | | wb_tint | 0.0 | Green ← → Magenta |

Global Tone

| Parameter | Default | Description | |-----------|---------|-------------| | enable_tone | True | Linear-light tone mapping | | exposure | 0.05 | EV adjustment — iPhone meters slightly bright | | contrast | 0.15 | Midtone contrast around the 0.18 photographic pivot | | shadows | 0.35 | Luminance-masked shadow lift + tiny black-point lift | | highlights | 0.5 | Reinhard soft-knee rolloff — smooth, monotonic, never clips |

Smart HDR

| Parameter | Default | Description | |-----------|---------|-------------| | enable_smart_hdr | True | Local tone mapping in log-luminance | | hdr_strength | 0.35 | Local dynamic range compression amount |

Color Science (Oklab)

| Parameter | Default | Description | |-----------|---------|-------------| | enable_color | True | Perceptual color rendering | | vibrance | 0.4 | Smart saturation — boosts muted colors more, hue-preserving | | skin_protection | 0.7 | Shields skin from vibrance + caps skin chroma (no orange faces) | | shadow_tint | 0.25 | Subtle cool/blue shadows — the iPhone blue-black signature | | highlight_warmth | 0.2 | Golden warmth in highlights |

Detail

| Parameter | Default | Description | |-----------|---------|-------------| | enable_detail | True | Two-scale luminance detail, halo-suppressed | | texture | 0.35 | Fine scale (~1px): pores, fabric, hair | | clarity | 0.2 | Mid scale (~2% of frame): local punch |

Optics & Sensor

| Parameter | Default | Description | |-----------|---------|-------------| | enable_optics | True | Vignette + grain | | grain | 0.2 | Photon-weighted noise, stronger in shadows | | vignette | 0.25 | Natural falloff in linear light — iPhone is well-corrected, keep low |

---

🎨 LUT Nodes

Two dedicated nodes for applying .cube 3D LUTs with tetrahedral interpolation (the industry-standard method used by Resolve and camera ISPs — more accurate than trilinear, especially on the neutral axis).

Recommended settings

I recommend setting the strength to around 0.05 to 0.10

[🎨 LUT Loader] → [🎨 LUT Apply] → [SaveImage]
                        ↑
                    [Your Image]

🎨 LUT Loader

| Parameter | Description | |-----------|-------------| | lut_name | Dropdown of .cube files from the luts/ folder |

🎨 LUT Apply

| Parameter | Default | Description | |-----------|---------|-------------| | strength | 0.85 | 0.0 = original, 1.0 = full LUT |

Bundled LUTs

| File | Type | Correct input | |------|------|--------------| | AppleLog2_to_Rec709_33_Grid.cube | Technical conversion | Apple Log footage only — do not apply to regular sRGB images | | Presetpro - Portra 800.cube | Creative film look | Regular sRGB images |

Drop your own .cube files into luts/ and restart ComfyUI.

---

📋 Requirements

• ComfyUI (latest version recommended)
• Python 3.10+
• numpy and torch (already included with ComfyUI)

🤝 Contributing

Pull requests are welcome — new Photographic Style tunings, LUTs, or pipeline improvements.

📄 License

MIT License — free to use, modify, and distribute.