Launching Reve 2.1

We are releasing Reve 2.1.

Just a month after Reve 2.0, our new model makes a rapid leap in visual intelligence and world knowledge.

Reve remains both the top 4k model in the world and the highest-ranked independent foundation image model lab on both Arena and Design Arena. The model takes back our spot as second overall, trained at less than a tenth of the compute of the labs ranked above and below us.

Reve 2.1 extends the same foundation of controllability, but sharpens it at every stage with intuitive prompt understanding and stronger foreign-text rendering. The model plans in finer detail, reasons more accurately about how elements relate, and renders the result with greater precision at native 4K.

• Visual intelligence and layout planning. The model reasons about structure, hierarchy, and spatial relationships before it renders, excelling at dense and complicated scenes.

• Precision editing. Every element is addressable and editable, so a single region can be changed and re-rendered

• Pristine 4K output. Renders at a full 16-megapixels, preserving every sharp edge and texture through iterative changes.

• Multilingual text rendering. Dense, legible type set inside the image, now extended to foreign scripts.

Reve is an independent lab training with efficiency, conviction, and focus. Two years ago, our team bet that images should be built like code, with layouts: hierarchical, structured regions. Through this bet, our models build controllability into visual intelligence by design. Reve 2.1 sharpens that control with a model that plans layouts more precisely than ever.

For more details on our research, read about Reve's large layout model and the extended reasoning context it enables in The Layout Bet.

Create, edit, and iterate freely with Reve 2.1 now at reve.com.