About VibeDex

Vendor bias everywhere

Most comparison sites sell API access to the models they rank, creating inherent conflicts of interest.

One-dimensional rankings

A single “quality score” hides the nuance. The best model for product photography is not the best for concept art.

No methodology transparency

Cherry-picked examples and vague criteria make it impossible to trust or reproduce results.

Multi-Dimensional Scoring

Models are evaluated across multiple quality dimensions — visual fidelity, physical accuracy, subject integrity, and prompt adherence — so comparisons reflect what actually matters for your use case.

Intent-Aware Matching

VibeDex analyzes your prompt to understand what it demands — then weights scores accordingly. A portrait prompt prioritizes anatomy; a product shot prioritizes realism and lighting.

Hybrid Evaluation

Our proprietary automated benchmarks are supplemented with public data sources, editorial research, and community reviewer feedback to ensure scores reflect real-world performance.

Fully Independent

We have no commercial relationships with model providers. Every recommendation is based on benchmark evidence, not sponsorship deals.

Automated Benchmarks

AI-powered visual judges score every generated image across multiple quality dimensions, calibrated for consistency and cross-validated across evaluation runs.

Public Data Integration

We supplement internal benchmarks with established public sources — industry leaderboards, editorial reviews, and published datasets — to validate and broaden our coverage.

Community Review

Human reviewers provide qualitative feedback on model outputs, validating automated scores against real-world creative standards and use-case expectations.

Visual Fidelity

Overall image quality and visual appeal, scored across:

• •Aesthetics — artistic quality, color harmony, visual impact
• •Image Quality — sharpness, noise, artifact-free rendering
• •Composition — framing, balance, visual hierarchy

Physics & Logic

Realistic lighting, materials, gravity, and physical plausibility. Scored across multiple sub-metrics targeting static and dynamic realism.

Subject Integrity

Accurate human anatomy, object coherence, and scene consistency. Sub-metrics evaluate subjects, objects, and spatial relationships independently.

Instruction Adherence

How faithfully the output matches the prompt — including semantic accuracy, spatial composition, and text rendering quality.

AI Evaluation Engine

Gemini 3 Pro serves as our primary VLM judge, processing 3,500+ evaluations across 20 models. We built a custom scoring framework with 4 quality dimensions and 12 sub-metrics for granular, reproducible assessments.

Intent Analysis

Our intent router uses LLMs to analyze user prompts in real-time, determining which quality dimensions matter most and dynamically weighting model scores to the specific use case.

Cloud Infrastructure

Built on Google Cloud — Cloud SQL for evaluation data and model profiles, Cloud Storage for generated image assets, and serverless compute for our API and batch evaluation pipelines.

Recommendation Engine

A custom scoring engine that combines intent-weighted quality scores with cost normalization, confidence penalties, and category-specific evaluation data to produce personalized model rankings.

API Access

Programmatic access to evaluation data, model profiles, and our recommendation engine for teams integrating model selection into their workflows and pipelines.

Team Plans

Dashboard access for creative teams to benchmark models against their own prompts and use cases, with saved preferences and collaboration features.

Enterprise

Custom benchmarking for organizations evaluating models at scale, with dedicated evaluation runs, private leaderboards, and integration support.