Reve Image 1.0
The 'Halfmoon' model that topped the image arena — elite prompt adherence and typography, from scratch.
About
Reve Image 1.0 is the debut text-to-image model from Palo Alto startup Reve AI, launched in March 2025 under the mysterious codename 'Halfmoon' — it climbed to #1 on the Artificial Analysis Image Arena (ELO ~1167, 100,000+ votes) before anyone knew who built it. Unlike most competitors fine-tuned on existing bases, Reve was trained from scratch to excel at three things: exact prompt adherence, aesthetics, and typography — reliably rendering readable text, logos, and signs where rivals garble them. It supports conversational editing ('change the sky to dusk'), reference-image remixing, and full commercial rights on outputs.
The honest picture: at launch there was no public API and no announced long-term pricing — it lived as a free preview, so production dependence was risky. Transparent materials and specific fictional characters were known weak spots. By 2026 the field caught up (Nano Banana, Seedream, Qwen-Image all press hard on the same strengths), but Reve remains a landmark release and a strong pick for text-heavy creative work.
What it can do
- Text-to-image with best-in-class prompt adherence
- Accurate in-image typography and text rendering
- Conversational image editing in plain language
- Reference-image remix (up to 6 images)
- Full commercial rights on generated outputs
Setup
Sign up on the Reve platform (preview.reve.art at launch) with an account; generate from text or upload a reference image and edit conversationally.
Prerequisites
A Reve account; API access was not publicly announced at launch.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 54; Trending score 78/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.