Backflip AI has added a way to turn flat engineering drawings into editable 3D models. According to the company's press release, dated September 17, 2026, the new Drawing to CAD tool builds parametric models with editable feature trees. It is live now on the Backflip web app and as an Autodesk Fusion add-in. A multi-photo Photo to CAD tool arrives alongside it, joining the existing Scan to CAD.
The announcement was made at Autodesk University. For makers, the interesting part is not the AI label but the output format. A parametric model with a feature history is a different object from a mesh, and that difference decides whether a converted part is something you can revise or merely something you can print.
Three Inputs, One Goal
Backflip now offers three routes into CAD, each starting from a different kind of source material.
- Drawing to CAD takes a 2D engineering drawing and produces a parametric 3D model with an editable feature tree.
- Photo to CAD reconstructs a part from 3 to 6 photos plus one known dimension, according to the press release. TCT Magazine describes the input as mobile phone photos.
- Scan to CAD takes 3D scans and meshes or STL files. TCT notes this tool already existed before the launch, and that Text to CAD was launched previously.
The single known dimension in the photo workflow is worth dwelling on. Photographs carry no inherent scale, so the tool needs one measurement to anchor everything else.
Why the Feature Tree Matters
Anyone who has pulled an STL of a scanned bracket into a CAD package knows the problem. The geometry is there, but it is a shell of triangles. Moving a hole, changing a boss diameter or adjusting a bolt pattern means rebuilding the feature by hand or hacking at the mesh.
A feature tree is the record of how a part was constructed: sketches, extrusions, holes, patterns. Backflip says Drawing to CAD delivers that structure. TCT's reporting, by Sam Davies, says engineers can inspect the design history in Fusion and then edit it as required. The press release spells out the kinds of edits: changing a hole diameter, adjusting a pattern or revising a mounting interface.
That is the practical payoff. If a legacy part has a worn mounting interface, you can change the interface rather than trace around it. Editing a dimension and re-exporting is the ordinary workflow for parametric CAD, and it is what mesh-based reconstruction has struggled to offer.
The press release adds that the tool can export STEP files for use in other CAD systems, and that more CAD integrations are in development. Anyone working in another package will need to go through STEP for now, which is a standard interchange route but one that the announcement does not describe in detail.
The Cost Claim, and a Small Discrepancy
Backflip's headline number is economic. TCT reports that reconstruction which typically takes 2 to 3 days and costs about $1,500 is claimed to finish in about 30 minutes for $10 or less. The press release frames it similarly: about $10 or less per conversion versus about $1,500 traditionally, with time of 5 to 30 minutes depending on complexity.
The two accounts differ slightly on time. TCT says about 30 minutes; the press release gives a range from 5 to 30. The range is the more specific statement, and it is consistent with TCT's figure being the upper end. Either way, both are company claims. Neither source describes an independent benchmark, and neither reports accuracy figures, so how faithfully a given drawing is reconstructed is not something the announcement lets us judge.
The comparison baseline is also worth reading carefully. The $1,500 and 2 to 3 day figures describe a traditional reconstruction, presumably done by a human drafter or engineer. A hobbyist rebuilding a knob in an evening is not the target of that comparison, but the per-conversion price still matters at the small end. At $10 or less, trying a conversion costs far less than the hours it would take to model a complex part from scratch.
Who Is Behind It
Backflip AI is led by founder and CEO Greg Mark, who also founded Markforged. The company is backed by NEA and Andreessen Horowitz. The press release quotes Mark on the problem the product is aimed at: "Many factories have 3D CAD for less than 1% of their equipment." He adds that when a part breaks, which he says happens often, there is no 3D model immediately available and it becomes a scramble to keep a production line from going down.
That framing points at industry rather than the garage. TCT reports that Backflip aims to unlock the knowledge trapped within legacy 2D drawing archives, and digitizing those archives is the market the company is describing.
What It Means for Makers
The most direct use case is reverse-engineering legacy parts for printing. Consider the situations where this matters:
- You have a manufacturer's drawing of a discontinued part and want a printable, adjustable model.
- You have a physical part with no drawing, and can photograph it and measure one dimension.
- You have a scan or mesh and want to convert it into something editable rather than a fixed shell.
In each case the value is the same: a model you can tweak for print. Printed replacements rarely match the original perfectly. Tolerances shift with material and process, and a parametric model lets you nudge a hole diameter or a clearance and reprint without starting over.
Some caution is warranted. The announcement is a launch, and the evidence offered is the company's own. Before relying on a converted model for a load-bearing part, check its dimensions against the original drawing or the physical object. Drawings can also carry ambiguities, such as notes, tolerances and views that do not agree, and no source here says how the tool handles them. The sensible approach is to treat the output as a strong first draft and verify the critical features yourself.
Still, the direction is clear. Turning drawings, phone photos and scans into editable, history-bearing models inside a tool many makers already use lowers the barrier to repairing and reproducing old hardware. If the quality holds up in independent testing, that is a meaningful shift for anyone who has spent an evening rebuilding a part from measurements.
