Introduction
This is a project to create a reproduction of a toki adze – a ceremonial axe from Mangaia in the Cook Islands – currently held in the Royal Museums Greenwich archive collection. The toki is approximately 850mm in length, with an intricately carved shaft of repeating cross-cut motifs, a polished dark stone blade, and fine coconut fibre lashing binding blade to haft.
The project sits within maker-identification research into Mangaian toki, undertaken with colleagues at Royal Museums Greenwich and connecting to work by carver Robbie Teremoana Atatoa.
Hundreds of Mangaian toki were removed from the Cook Islands during the nineteenth century and now exist in in museum collections around the world. While preserved physically, they are largely inaccessible to the communities and carving traditions they came from. This project explores whether modern photogrammetry and additive manufacturing can produce a faithful physical reproduction suitable for handling, study and cultural reconnection.
Photogrammetric Capture
This toki was captured over a supervised session in the RMG archive store, working with museum colleagues. Capture setup was tripod / DSLR-based photogrammetry using a Canon EOS R6 Mark III with a 24–105mm lens, shooting RAW at f/8–f/11 with manual focus and image stabilisation off.
The object sat on protective foam on a turntable, under LED studio lighting. White balance locked in to keep colour consistent across the full image set. The shot strategy was 36 frames per loop (every 10°) across four elevation passes; around 144+ frames for each side side.
Since the toki could not be stood upright – or readily handled – it needed to be photographed and flipped to capture each side separately. This breaks the geometric continuity that photogrammetry relies on, so the two sessions were essentially separate reconstructions from the outset rather than trying to force them into a single alignment.
Reconstruction and the Merge Problem
The image sets were processed with RealityCapture into dense meshes. The two flipped sessions reconstructed very well individually but as separate components – and with a slight scale discrepancy between them of roughly 1.1x – a known issue of turntable photogrammetry where the software has no absolute scale reference.
An attempt to resolve the registration externally in CloudCompare didn’t succeed, and the merge was instead worked through inside RealityCapture itself – placing manual control points across features visible in both sessions and iterating until the two halves pulled into agreement. The final alignment was close but not quite perfect – a fraction of residual mismatch that had to be accepted and tidied at the mesh stage rather than eliminated at source.
It’s a real reflection of flip-scan photogrammetry on an object like this: with no shared geometry between the two sessions beyond the object itself, there is a practical limit to how exactly two independent reconstructions can be forced together.
[note for future version: investigate small object laser scanner]
Limitations of Photogrammetric Reconstruction
Photogrammetry captured the overall form and the colour texture very well, but it has known limitations: sharp edges soften and deep dark crevices don’t reconstruct as they’re too much in the dark. On this toki, that meant the deep V-cut channels separating the 45 carved rings along the shaft came through too shallow – one of the most striking features of the object, lost just where the tool marks cut deepest.
To address this the geometry was brought into Blender 3d modelling software for some digital sculpting to reintroduce features visible on the original but missing from the mesh; sculpt mode sharpening to recover some edge definition and a series of torus rings Boolean subtracted from the shaft to recut those grooves.
This is the point it becomes clear this project has edged from replication / documentation into interpretation – an informed reconstruction rather than a pure scan copy.
Some sculpting back of the lashing zone on the head at this stage too – to make space on the finished print for real cordage, rather than a printed imitation of fibre binding which would never look authentic.
Fabrication
The reproduction was printed on a Bambu Lab H2D. At 850mm the toki far exceeds the 325mm build volume, so the handle was split into three sections in Blender, with the joints planned to fall where they could be disguised within the carved pattern.
PLA Wood Black Walnut and Rosewood – wood-filled filaments whose matt, slightly fibrous surface takes wood dye convincingly. The head stone was printed separately in a dark technical filament to read as stone against the timber tones of the shaft.
Print orientation took a test print to get right. The instinct was to print the handle sections lying down, but the results read badly: with the shaft on its side, the upper and lower surfaces of the cylinder meet the build plane at very acute angles, so each layer covers a wide sweep of the curve, and the stepping between layers becomes strongly evident – the carving dissolves into visible contour bands. Printed standing up, the geometry meets the layers far better: each ring builds as a series of near-horizontal cross-sections, and the V-cut channels keep more definition. A relatively coarse layer height (0.25–0.28mm) was retained deliberately; layering can work with a wooden object rather than against it.
Assembly
The three handle sections were joined permanently with cyanoacrylate adhesive – glue on one face, activator mist on the other, dry-fitted first, then held firm on contact. CA bonds well to PLA specifically, keying into the layer lines. Glue squeeze-out was kept clear of show surfaces, since cured CA resists dye and telegraphs through wax as a sheen difference.
Assembling before finishing was necessary as the shaft is dyed as one continuous piece, so the tonal gradient – richer and darker toward the head, lighter along the handle can run unbroken across the joints.
Finishing: Building Aged Wood in Layers
The finishing pipeline treats the print as if it were timber, sanding then using actual wood-finishing products rather than just paint effects.
An early lesson: water-based stain beads on smooth PLA. The answer was to work with dyes that penetrate the wood-filled surface – Liberon water-based wood dye as the foundation, with solvent-based spirit dye in a darker oak building depth. The first coat is applied diluted, which runs naturally into the recesses and settles pigment exactly where age would deposit it; subsequent coats build the overall tone.
[note for future version: solvent based dye to darken light coloured filament]
The worn, handled character of the original – lighter peaks where generations of hands have polished the carving – was reproduced by working a nearly-dry cloth of lightened tone (an ochre-tempered mix) across the peaks only, not into the grooves.
The final layer is Liberon Black Bison paste wax in Antique Pine – buffed hard on the peaks for the soft sheen of handled wood, and just lightly into the recesses with a toothbrush so the grooves keep their matt depth.
Lashing
The lashing is where the digital workflow stopped. The coconut fibre binding on the original is not a surface texture but a structural craft in its own right – a specific sequence of wraps and crossings that no 3D printing process can properly replicate. The binding on the reproduction was undertaken by Robbie Teremoana Atatoa, the only member of the project with the skills to do it correctly.
This means the reproduction is not wholly a machine translation of the original: its binding was made by hand, in the proper manner, by a carver whose practice descends directly from the tradition the toki belongs to. Where every other stage of the workflow is a copy of the object, the lashing is a continuation of it.
The finished toki is not the original. What it tests is how much of an object’s character – its wear, its tool marks, its accumulated handling – can survive translation through scan data and extruded filament. And unlike many of our reproductions, this one has somewhere to go: it travels with Robbie to New Zealand, a faithful, handleable copy within reach of the carving tradition it came from, while the original remains in the archive in Greenwich.
Credits
- Scan photography undertaken with access to the original toki kindly facilitated by Royal Museums Greenwich: Ceremonial toki from Mangaia, Cook Islands (ZBA5519).
Royal Museums Greenwich | Home - Equipment for this project was purchased via the ICE Collaborative Research Development Fund 2025/26
Institute for Inclusive Communities and Environments - Project lead and cultural research: Robbie Atatoa. This recreation was undertaken in support of Robbie’s ongoing work to identify, translate, recreate and revitalise Mangaian toki, developed through his Churchill Fellowship research.
Robbie Atatoa
