1.2Comparison Approaches

Approach	How It Works	Strengths	Weaknesses
Pixel diff	XOR each pixel, highlight differences	Deterministic, fast, no false negatives	Extreme false positive rate
Perceptual diff (pdiff)	Model human vision sensitivity to changes	Fewer false positives than pixel diff	Still trips on font rendering, anti-aliasing
Structural similarity (SSIM)	Compare luminance, contrast, structure	Better at ignoring compression artifacts	Cannot understand layout semantics
DOM-aware diff	Compare DOM structure + computed styles	Ignores rendering engine differences	Misses visual-only bugs (z-index, opacity)
Vision model (AI)	Send screenshots to a multimodal LLM or specialized model	Understands intent, ignores noise	Slower, costs per comparison, non-deterministic
Hybrid (modern tools)	Pixel diff first, AI triage for flagged changes	Fast for unchanged screens, smart for changed ones	Complexity in pipeline setup

The Hybrid Approach in Detail

The most effective strategy combines fast deterministic checking with AI intelligence:

1. Pixel diff first -- fast comparison flags any screenshots with differences
2. Threshold filter -- changes below a configurable pixel ratio (e.g., 0.1%) are auto-approved
3. AI triage -- changes above the threshold are sent to a vision model for classification
4. Human review -- only genuinely meaningful changes require human attention

This reduces the number of screenshots requiring human review by 80-90% compared to pure pixel diff, while maintaining near-zero false negative rates.