Reve released Reve 2.0 on June 3; the layout-based AI image generator ranks #2 on the Arena text-to-image leaderboard, behind GPT Image 2 and ahead of Nano Banana 2. Reve 2.0 renders images at native 4K resolution, approximately 16 megapixels, and maintains the original model’s low per-image pricing, with API generations costing around a fraction of a cent per image. The model uses a layout-based approach that enables targeted edits without re-rolling the entire picture.
Reve 2.0 was trained using ten times fewer GPUs than the larger models it sits beside. The model renders images at native 4K resolution, equivalent to about 16 megapixels. API generations remain inexpensive, costing around a fraction of a cent per image, continuing the original Reve model’s low-cost approach that previously beat Midjourney and Flux at roughly a cent per image. Eight areas were tested to evaluate the model.
Reve 2.0 uses a layout-based approach that allows selective modifications such as moving a subject, rewriting a sign on a wall, or swapping a background without re-rolling the entire picture. The layout is a structured, editable description in which objects have specified locations, sizes, and captions, a format compared in the source to how HTML structures a webpage. Reve’s image quality is described as filmic, with a photojournalistic look, and it is less glossy than Nano Banana 2 and GPT Image 2, which still have an edge in pure realism. If prompts are overly long, the model can struggle with many details, although in certain tests Reve 2.0 still outperformed GPT Image 2.
These points summarize Reve 2.0’s documented technical and operational features. They reflect the model’s training footprint, cost profile, editing approach, and reported image character.
Eight areas were tested to evaluate Reve 2.0. The evaluations found that when prompts are too long the model can struggle with many details. Despite those struggles, Reve 2.0 can still outperform GPT Image 2 in certain tests. These findings were reported across the eight tested areas.
“If you iterate heavily, care about text, print at high resolution, or build agentic pipelines, then the layout approach is a real edge.” The layout approach allows moving a subject, rewriting a sign on a wall, or swapping a background without re-rolling the entire picture.
The evaluation therefore highlighted both the model’s strengths in targeted editing workflows and its limitations with overly long, detail-heavy prompts.
Reve 2.0 ranks #2 on the Arena text-to-image leaderboard and pairs cost-efficient API image generation with native 4K output suitable for high-resolution applications. Its editable layout feature enables targeted, in-image modifications without full regeneration, supporting workflows that require precise edits and high-resolution results while maintaining a low-cost production profile overall.
