Asalah

We see AI-generated images of Saudi culture every day, and most of us can't tell they're synthetic. There's no public benchmark that measures how culturally accurate these models actually are, or how well people can spot the difference. Asalah (أصالة) — Arabic for "authenticity" — is a community-driven platform that puts both questions in one place: a literacy game that trains the eye, and an arena that turns the crowd's votes into a leaderboard.

What we're building

🔍 Spot the Real AI literacy Each round shows one real photograph of Saudi culture beside an AI-generated image of the same scene. Pick the real one. A timed mode adds live, hosted group sessions.

⚔️ Cultural Arena Model battle Two AI-generated images, same cultural prompt, different models, shown fully blind. The crowd votes; the votes feed a live leaderboard ranking models by cultural fidelity, not general image quality.

Everything visitors see is pre-generated ahead of time — curated cultural scenes are paired with a verified real reference photo, candidate models generate for that exact scene in advance, and only then does a visitor compare and choose. Choices become data; data becomes the leaderboard.

Five Saudi cultural categories

Why the leaderboard holds up

• ~350 generation prompts across the five categories, each paired with a verified real reference photo
• Few-shot conditioning — models are graded against real reference images, not judged on text descriptions alone
• Bradley-Terry scoring with statistical confidence intervals, not a raw vote tally
• Saudi community voters, not crowdworkers with no stake in the culture being represented
• Open dataset — released openly once the benchmark launches, for anyone to build on

Built at Alfaisal University's AI Research Center

Asalah started under the SURE undergraduate research program at the AI Research Center (AIC), Alfaisal University, Riyadh, with reference photography and voters drawn from partner universities across the Kingdom.

Status

This is an early-stage research prototype, not a finished product: there's no live public deployment yet, generation is offline and pre-computed, and the leaderboard is still gathering enough votes to be meaningful. What's real already: the taxonomy, the prompt set, the scoring methodology, and a working literacy game and arena. Follow @Asalah_ai for the public launch, or reach the team at Asalahteams@gmail.com.