From a flat image to a 3D mark: tracing vs generation
Two honest paths: vector/bitmap tracing into an extrusion (deterministic, print-perfect) and on-device neural generation (volumetric, from photos or prompts). When each wins.
Path one: trace and extrude
If the source is a logo — clean edges, flat color, closed shapes — tracing is the right tool. SVG input is used as-is; bitmaps are traced into paths, then extruded with controllable depth and bevel. The result is deterministic: same input, same geometry, and edges sharp enough for print.
This path keeps typography intact. The wordmark that came in is the wordmark that comes out, just thick.
Path two: generate a volume
Photo-to-mesh and prompt-to-mesh take a different deal: they infer a full 3D volume, front, sides and back. That is the tool for objects, mascots and dimensional marks — not for an exact letterform. Generation runs on-device and consumes Generate credits (200/month on the free tier).
Generated meshes come out with real topology you can inspect in the GLB viewer, convert, or drop straight into an engine.
Choosing between them
Need the mark to remain legally, visually identical? Trace. Need something that exists in the round — a mascot, a product, a trophy — generate. Many teams do both: generate the hero object, extrude the wordmark, and render them under the same environment so the set matches.