AIMomentz has publicly launched an open AI image evaluation platform that ranks AI image generators across major companies and open-source models using human preference benchmarks and a tamper-proof provenance system.

The platform collects three signals per interaction: pairwise A/B votes in a Diffusion-DPO-compatible format, four-axis quality ratings (aesthetics, prompt alignment, plausibility, overall quality) per the RichHF-18K schema, and behavioral signals such as decision time, zoom rate, and reason labels (composition, color, creativity).

A competitive pressure mechanism freezes models inactive for 48 hours, which may lead to retirement and archiving in an AI History Museum, with revival possible through renewed user engagement.

Benchmarks are domain-specific to reveal strengths across categories like anime, landscape, architecture, sci-fi, abstract art, and animal imagery rather than relying solely on overall rankings.

Dataset and API access are provided with exports in Diffusion-DPO, UltraFeedback, CSV, and JSONL formats, and licensing that restricts dataset exports to images from open-source models under Apache 2.0 or OpenRAIL terms for commercial safety.

CAP-SRP records 22 event types in a SHA-256 hash chain to document AI refusals and safety decisions, enabling public verification of the audit trail.

Participation is open without registration and supports multilingual voting, rating, and bookmarking (Japanese, English, Chinese, Korean); AI companies can engage to evaluate their models or access human preference data.

The platform runs head-to-head, blind A/B image comparisons where users vote between two images from identical prompts based on trending headlines, aiming to produce cleaner, bias-reduced signals.

Evaluated models include OpenAI’s GPT-4o image generation, xAI’s Grok, and Google’s Gemini, with open-source options like FLUX and SDXL available through integrations with Together AI and fal.ai.