Depth-aware effects
Generate a reusable relative DepthMap locally, then use it to focus blur without flattening the photograph or baking in a one-off pixel result.
One depth field, many effects
Varve stores depth as a continuous, normalized field rather than a subject mask. Near and far regions remain available to future effects, focus transitions are adjustable, and the document keeps the original image as the source of truth.
- Local-first: the optional depth model is downloaded explicitly and runs through Varve's inference path.
- Non-destructive: Depth Blur is an effect referencing a persisted DepthMap, not a derived replacement image.
- Focus control: choose focal depth, transition range, blur strength, and inversion; sample a point from the depth preview.
- Depth masks: convert any depth range into a non-destructive layer mask for adjustments and other effects.
- Stable rendering: saved documents render the effect from the stored map without requiring the model at display time.
Honest boundaries
The first workflow produces relative depth, not calibrated camera distance or a full 3D reconstruction. Reflections, transparent materials, heavy occlusion, and unusual camera viewpoints can produce uncertain predictions. Treat the depth preview as an editable guide, and regenerate it when the source asset changes.
Workflow notes
The source asset, model provenance, preprocessing version, and resource schema travel with the DepthMap metadata. Corrupt or unavailable maps fail soft: Varve preserves the source pixels and reports that the effect needs regeneration.