Check a face photo for Photoshop- or Facetune-style warping
Geometric edits only. Not makeup. Not AI-generated faces.
ICCV 2019
Geometric warp
Not a deepfake detector
1. Input Image
Upload a clear JPEG or PNG portrait (Max 6 MB).
Click or drag & drop face photo here
JPEG or PNG up to 6 MB
2. Analysis Output
Detection output slots (v1 placeholder state).
Slot A: Warp Score / Signal Tier
Awaiting input image...
FAL global classifier: chance this face crop looks like Face-Aware Liquify / similar warp. Not percent the photo is fake.
Slot B: Heatmap Visualization
Heatmap visualization will generate here.
Slot C: Suggested Geometry Undo
Suggested geometry restoration preview will render here.
Reference Output Architecture
Visual model examples demonstrating what FAL produces during deep analysis:
Face Crop (Standardized)
1. Cropped Face
Normalized facial bounding box centered on landmarks.
Localized Heatmap Gradient
2. Liquify Heatmap
Per-pixel flow field highlighting localized displacement vectors.
Inverted Warping Estimation
3. Suggested Undo
Estimated reverse vector field to approximate original geometry.
Key Constraints & Limits
- Needs a visible face: Algorithm relies on facial landmark calibration.
- Geometric drag only: Specifically tuned for Photoshop Liquify, Facetune reshape, and equivalent tools.
- Misses skin smoothing & lighting: Retouching, frequency separation, and color grading are not measured.
- Web compression degradation: Heavy JPEG artifacts or repeated social media re-compression reduce precision.
- Signal, not a verdict: Low score ≠ honest photo; High score ≠ proof a named person edited it.