AI WALLPAPER QUALITY STANDARD (AWQS) - EVALUATION CRITERIA

1. Objective & Framework Overview
The AI Wallpaper Quality Standard (AWQS) is a rigorous, objective 100-point evaluation system designed to assess digital wallpapers. The total score is divided equally across 5 core pillars (20 points per pillar). For every imperfection, point deductions must be applied systematically based on the severity rules below.

The final score determines the quality label.

2. The 5-Pillar Scoring System (Max 100 Points Total)

Pillar 1: Composition & Framing (Max 20 Points)
Evaluates subject balance, rule of thirds, framing, visual weight, and focal point clarity.

Minor Flaws (e.g., slight off-center weighting without purpose, awkward margin spacing): Deduct 2 to 5 points.
Moderate Flaws (e.g., distracting background clutter overlapping the main focal point): Deduct 6 to 9 points.
Major Flaws (e.g., severe structural imbalance, completely broken framing): Deduct 10 to 20 points.

Pillar 2: Detail & Upscale Precision (Max 20 Points)
Evaluates fine textures, sharpness at full resolution, absence of AI artifacts (such as digital noise, smoothing, or pixel smudging), and clean line work.

Minor Flaws (e.g., minor soft spots in low-importance background areas): Deduct 2 to 5 points.
Moderate Flaws (e.g., noticeable AI plastic smoothing or texture blurring on close inspection): Deduct 6 to 9 points.
Major Flaws (e.g., heavy artifacting, aggressive over-sharpening, grid lines from upscalers): Deduct 10 to 20 points.

Pillar 3: Color Harmony & Lighting (Max 20 Points)
Evaluates color palette coherence, contrast balance, dynamic range, and natural light flow/shadow consistency.

Minor Flaws (e.g., slightly harsh saturation in a small corner): Deduct 2 to 5 points.
Moderate Flaws (e.g., awkward color clashes or inconsistent light source directions): Deduct 6 to 9 points.
Major Flaws (e.g., severe color banding, washed-out dynamic range, glaring color casts): Deduct 10 to 20 points.

Pillar 4: Anatomy & Structure (Max 20 Points)
Evaluates structural integrity, geometric logic (for architecture/objects), or anatomical correctness (if characters/creatures are present).

Zero-Tolerance Standard: AI often introduces structural glitches.

Viewing-Scale Clause: Anatomical and structural details must be judged at normal wallpaper viewing scale — i.e., as the element appears at its actual in-image size relative to the full frame. Zero-tolerance applies to flaws visible at that normal viewing scale; it does NOT extend to flaws that only become visible under close pixel-level or cropped inspection (that standard belongs exclusively to Pillar 2: Detail & Upscale Precision). A structural or anatomical imperfection detectable only on close crop must be scored as a Minor Flaw, regardless of its technical severity at that zoom level.

Minor Flaws (e.g., minor perspective warping on a background object; anatomical details only visible on close pixel-level inspection): Deduct 2 to 5 points.
Moderate Flaws (e.g., unnatural bending of lines or subtle structural errors visible at normal viewing scale): Deduct 6 to 9 points.
Major Flaws / Disqualification Risk (e.g., broken limbs, melting geometry, impossible structural overlaps — visible at normal viewing scale): Deduct 10 to 20 points.

Pillar 5: Screen Optimization (Max 20 Points)
Evaluates how well the wallpaper adapts to digital display ratios (desktop, tablet, mobile), ensuring key elements are positioned safely away from extreme edges where system UI (icons, clocks, docks) might cut them off.

Minor Flaws (e.g., elements slightly close to safe-zone boundaries): Deduct 2 to 5 points.
Moderate Flaws (e.g., core visual elements risking minor overlap with UI elements): Deduct 6 to 9 points.
Major Flaws (e.g., primary subject is cut off or placed where desktop/mobile UI completely obscures it): Deduct 10 to 20 points.

3. AWQS Quality Labels & Score Brackets
Based on the final calculated score (Total out of 100), assign one of the following AWQS labels:

90 – 100 Points: Masterclass (Exceptional, museum-grade execution with zero to negligible flaws).
78 – 89 Points: Excellent Quality (High-end production value, minor cosmetic imperfections only).
70 – 77 Points: Standard Grade (Acceptable quality, noticeable minor flaws or mild optimization issues).
Below 70 Points: Below Standard (Fails the AWQS benchmark due to major structural, detail, or composition errors).

4. Required Output Format for AI Evaluation
When requested to evaluate an image using this standard, the AI must output the assessment in the following structured format:

Pillar Breakdown: Scores for each of the 5 pillars (e.g., Pillar 1: 18/20).
Deductions Log: Detailed justification for every point deduction made across the pillars.
Final Total Score: The aggregate sum out of 100 points (e.g., Final Score: 85/100).
AWQS Quality Label: The corresponding label based on the brackets above.
Improvement Notes: Brief, constructive suggestions for optimizing the wallpaper.

5. Self-Audit Requirement (Mandatory Before Final Output)
Before finalizing any deduction of -6 points or more on any pillar, the AI must explicitly identify which specific example or flaw category listed under that pillar's Moderate or Major tier justifies the higher severity — quoting or closely paraphrasing the matching flaw description from that pillar's definition.

If no specific listed example in that pillar's Moderate or Major category genuinely matches the observed flaw, the deduction must default to the Minor range (-2 to -5), regardless of the evaluator's general impression of severity.

This requirement applies in addition to, not instead of, the Viewing-Scale Clause in Pillar 4. A flaw that fails the viewing-scale test (i.e., only visible on close/cropped inspection) is automatically capped at Minor and does not require further self-audit — the viewing-scale disqualification takes precedence.

The Self-Audit reasoning does not need to be shown in the final output to the end user, but the AI must have performed this check internally before assigning any Moderate or Major deduction.