A Blender add-on that turns flat photos into surface geometry

What it does

DeepBump is a Blender add-on that uses machine-learning models to synthesize normal maps, height maps, curvature maps, and upscaled textures from single images. You select an image node in Blender’s Shader Editor and click a button; the add-on processes the texture and injects the result back into your material graph. A command-line interface is available for batch work outside Blender.

The interesting bit

The add-on keeps the entire workflow inside Blender’s Shader Editor, turning what is usually a multi-application round-trip—export to a standalone normal-map generator, process, re-import—into a single button click. The authors also explicitly wire up Microsoft’s telemetry-disabling APIs, a rare nod to privacy in a graphics utility.

Key highlights

• Generates normal, height, and curvature maps from a single albedo image.
• Upscales low-resolution textures without leaving Blender.
• Ships as a standard Blender add-on with a CLI mode for scripting.
• Disables Microsoft ONNX Runtime telemetry via available APIs.
• Released under GPL, though the training code remains unavailable.

Caveats

• Training code is not available, so the model is a black box you cannot retrain or audit.
• Dependency installation requires an internet connection and occasionally administrator privileges, which is a friction point on locked-down machines.

Verdict

Worth a look if you are a Blender artist or technical director who needs quick surface detail without round-tripping through external tools. Skip it if you need open, reproducible training pipelines or a standalone application outside the Blender ecosystem.