Reve's first image-editing model ranks top three
Image startup Reve says its first editing model ranks among the top three on LMArena and Artificial Analysis, which would put a 10-person research team alongside much larger labs in image editing.
Reve announced its first image-editing model. The company says the model ranks in the top three for image editing on both LMArena and Artificial Analysis, less than six months after Reve released a small research preview.
According to the company, the model handles single-image and multi-image edits and can combine several reference images into one output. Reve says it does especially well at removing shadows, editing landscapes and restoring images.
Reve says the model is designed to produce varied outputs. The company acknowledges this may lower its scores on benchmarks that measure average performance, but argues it gives users a better chance of getting the image they want. The post does not give exact benchmark scores or name the models ranked above or below Reve's.
The company stresses that its research team has only 10 people and says another model is "coming soon," without giving a date.
Reve also describes itself as a product company. It says its editor is the only one that lets users edit regions of an image directly, for example to move objects or recompose a whole scene.
Reve also announced new API pricing of 750 credits per dollar. The company says this works out to $40 per 1,000 images for remix and edit tasks and $24 per 1,000 images for plain text-to-image generation.
The post does not include details on the model's architecture, size or training data.