Some photo-prep tools market "real depth estimation, not AI guesswork" — and for their job, that is the right call. If the goal is reproducing *the photo you already have* as a height field, faithfulness is the whole point. But treating faithful estimation as the only honest way to make a relief misunderstands what relief work actually is.
Two different questions. Depth estimation answers: *how far was each pixel from the camera?* Relief sculpting answers: *how should this subject be carved?* A photograph of a face contains perspective depth — the nose is closer than the ears by centimeters. A minted coin of that same face compresses the whole head into two millimeters of designed relief: silhouette edges get crisp bevels, cheeks become smooth planned curves, the background becomes a flat polished field. No amount of faithful measurement produces that file, because it was never in the photo to begin with. Die engravers have made that translation by hand for centuries; AI Sculpt makes it in seconds.
When faithfulness wins:
- Portrait and memorial plaques — the customer expects the exact likeness, and material texture carries the depth illusion. - Lithophanes — the physics of backlit thickness demand tonal fidelity to the source. - Terrain and object scans — the data *is* the product; reinterpretation would be error.
For these, use the free deterministic depth engine: zero credits, no reinterpretation, true 16-bit.
When sculpting wins:
- Coins and medallions — literal photo depth reads waxy and shallow; minting relief is a designed compression. - Decorative brass and jewelry dies — you want beveled, domed, stylized planes, not camera geometry. - Logos and flat art — there is no camera depth to estimate; the relief must be invented.
The honest test for any depth tool is not "estimation vs AI" — it is whether the output carves clean. Ask three things: What bit depth does it actually output? (If unstated, assume 8-bit — see 8-bit vs 16-bit depth maps for why that bands.) Can you preview the carve in 3D before burning? And can you choose between faithful and sculpted for the job in front of you? Reticle Red is built to answer yes to all three: a free deterministic engine when the photo is the product, AI Sculpt when the carve is the product, both at true 16-bit with a live 3D preview.
Frequently Asked Questions
What is the difference between depth estimation and AI relief sculpting?
Depth estimation measures the photo you already have: it infers how far each pixel was from the camera and writes that as a height field, staying faithful to the source. AI relief sculpting designs a carve: it reinterprets the subject as a bas-relief — beveled edges, domed fields, simplified planes — the way a die engraver would, rather than the way a rangefinder would. Neither is "cheating"; they are different jobs.
When is a faithful depth map the right choice?
When the photo itself is the product: portrait plaques, pet memorials, photo lithophanes, and any job where the customer expects to recognize the exact image. Faithful estimation keeps proportions and likeness intact. Reticle Red ships a free deterministic depth engine for exactly this — zero credits, no AI reinterpretation, true 16-bit output.
When is sculpted relief the better choice?
When the output is a manufactured object rather than a reproduced photo: challenge coins, medallions, jewelry dies, and decorative brass panels. A literal photo depth map of a face makes a shallow, waxy coin — real minting relief compresses depth, sharpens silhouettes, and domes the fields. That is a design decision a faithful estimator will never make for you, and it is what AI Sculpt mode is for.
Does bit depth matter for both?
Yes — banding does not care how the heights were produced. Whether the height field was estimated or sculpted, 8-bit output terraces on deep carves. Both of Reticle Red's depth pipelines (deterministic and AI Sculpt) generate true 16-bit height fields end to end.