Shape, Photo, and Blend.
These three engines run entirely in your browser — no cloud call, no credits, no limit on generations. They cover the two fundamentally different ways an image can imply depth: shape and tone.
Shape
Built for flat 2D illustrations rather than photographs — logos, crests, line art, clean graphic designs. Shape uses a distance-transform approach: it reads the silhouette of your artwork and carves outward from the edges, rather than trying to interpret grayscale brightness as depth. That makes it the right choice whenever the source has sharp, readable shapes rather than continuous tone.
Photo
Reads grayscale brightness directly as height — brighter areas rise, darker areas recede (or the reverse, depending on your Relief Polarity setting). Detail Strength controls how much fine tonal texture survives into the height map, Denoise smooths sensor noise before it becomes surface roughness, and the Luminance Curve reshapes how brightness maps to depth so midtones can be pushed up or down independently of the extremes.
Blend
Mixes the Shape and Photo reads together in a single relief — useful when a subject is part flat graphic and part photographic tone, such as a logo composited over a photo background, or artwork that combines clean linework with shaded fill.
Need photorealistic depth instead?
Photographs of faces, figures, and organic subjects usually read better through the cloud AI Depth model than through Photo's flat luminance map.
See AI Depth →