Reve releases v1.5 image model with native 4K
Reve has released an early version of its v1.5 text-to-image model, which generates 4K images directly in pixel space rather than through a latent autoencoder, and says it already ranks in Arena's top three.
Reve announced an early release of Reve v1.5, which it calls its most capable text-to-image model so far. The company says the model already ranks among the top three on Arena. It did not name the leaderboard category, give a score or rank, or publish other benchmark results.
The main technical claim concerns how the model reaches 4K output. Reve says v1.5 works "in native pixel space, as opposed to the common practice of using a 'latent autoencoder' that loses precision." Most current diffusion-based image models generate in a compressed latent space and then decode to pixels. Reve says skipping that step keeps fine detail at full resolution and produces print-ready output. The post does not cover the architecture, compute costs or generation speed.
The company also says the model handles three things better than before:
- Detail and lighting: fine textures, complex scenes, and light and shadow.
- World knowledge: a better understanding of real places, objects, styles and concepts, so prompts are interpreted more accurately.
- Resolution: 4K output that stays sharp at full size.
Reve describes this as an early version and says image editing and reference-image inputs are coming soon. It gave no timeline. The post does not mention API access, pricing or licensing terms. The model is available to try at reve.com.