One picture can become a screen-only model, a game character or a printed figure, and each route needs different work. Here is the five-step photo-to-STL pipeline, the print settings a small figure needs, and how to convert an existing AI action-figure image into a real object.
A photo of a friend, a character sheet from an AI generator, a frame grabbed from a game: people bring all three with the same request, to make a 3D action figure out of it. Where the result ends up changes the job, because a web viewer, a game engine and a printer bed each reject a different kind of file.
This tutorial answers the questions that come up first: which of the three kinds of action figure you actually need, how a photograph becomes a printable STL, how an existing AI action-figure image becomes a real object, and what a generator will and will not hand you at the end.
Every number below is a typical range from common practice rather than a measurement from one specific test. Printers, resins and slicer profiles differ enough that the figures here are starting points, to be adjusted after the first test print.
Ask two people what a 3D action figure is and you get two different files. One wants a posed character that opens in a browser viewer or drops into a game engine. The other wants a physical object in the hand, printed in resin or FDM, standing on a base. A third only needs a still that reads as three-dimensional for a thumbnail. The geometry overlaps; the finishing work does not. Only the physical figure needs a closed volume, and only the game character needs a tight triangle budget and clean UVs.
| What you want | Final file | Typical route | Hands-on time |
|---|---|---|---|
| A 3D preview or turntable | GLB or glTF for a viewer or video | One image-to-3D generation, light cleanup | 10-30 minutes |
| A game or Roblox character | FBX or GLB, low-poly mesh, one texture set | Generate, then retopology and UVs in Blender | Half a day to two days |
| A figure for a shelf | A watertight STL or 3MF solid | Photo to mesh, repair, scale, hollow, slice | 3-6 hours plus print time |
| A painted collectible | The same solid, split into parts | Print route plus splitting, keys and primer | A weekend including paint |
Work backwards from the device that consumes the file. If the last stop is a slicer, every open edge and flipped normal becomes a failed print, so budget time for repair. If it is Unity, Unreal or Godot, the real cost is polygon budget and texture resolution. If it is a phone browser, download size decides whether anyone waits, and decimation plus compressed textures matter more than tidy topology.
The route from a photograph to a printable figure is short, but the order matters. Slicing straight after generation often works, which is the problem: when it fails, it fails after a resin tank is dirty and a hollow shell has closed around an invisible hole.
A generated mesh is a surface, not a solid: holes where the engine could not see the subject, self-intersections at the armpits and hair, duplicate shells inside the body. A slicer intersects that surface with horizontal planes, and the math needs every edge shared by exactly two faces to know what is inside. Most slicers repair small defects quietly, which is worse: the part prints with a thin patch that appears only when it cracks. A non-manifold selection plus a fill-holes pass catches most of it.
| Part of the figure | Resin (SLA/DLP) | FDM, 0.4 mm nozzle | Why it matters |
|---|---|---|---|
| Body and limbs | 1.5-2.5 mm wall after hollowing | Three perimeters, 1.2-1.6 mm | Thinner walls crack in resin and split along layer lines in FDM |
| Fingers, straps, weapon tips | 0.4 mm minimum feature | 0.8 mm minimum feature | Below that, the detail is not in the geometry |
| Overhangs | 0.03-0.05 mm layers, light supports | 0.12-0.2 mm layers, supports past 45 degrees | Support marks land on the face and hands |
| Base | Solid plinth, 3-5 mm | Solid, 4-6 mm, with a 3 mm brim | A 150 mm figure with a raised arm tips over |
An AI action-figure image is a good starting point: the character is already stylised, posed and lit the way the finished figure should read. What it is not is a model. One image gives an image-to-3D engine a single silhouette, so the back of the head, the far arm and anything hidden are invented rather than observed. That is why the same prompt run twice can return two different backs, and why generating two or three extra views to send as multiview input is the practical move.
Generator output is a starting mesh, not a print-ready part. No image-to-3D tool can promise a watertight solid from one photograph, and the file still needs repair, a scale check and a small test print. Engine availability, credit pricing and third-party terms change, so verify them on the vendor page first.
Open3DAI runs one local procedural generation engine and can also connect to third-party engine APIs with your own key. Useful here: image-to-3D and text-to-3D generation, multiview input, AI texturing, and side-by-side comparison of one picture across the available engines. Export covers GLB, glTF, OBJ, STL, PLY and 3MF, plus 48 format converters, 12 browser tools and a REST API.
What it does not do matters just as much. It does not run photogrammetry or scanning, rig or animate a character, retopologise or unwrap UVs automatically, or produce CAD-quality parametric solids, and it does not guarantee a watertight result with no cleanup. Those jobs still belong to Blender, ZBrush, Meshmixer, Fusion 360 and FreeCAD; a workable split is generator for the shape, desktop tool for precision.
Once one figure prints well, the reusable asset is the settings, not the model. Keep the photo checklist, the repair checklist and the slicer profile in one place. Name files with the character, scale and version, for example wren-1to12-v3.stl, so a reprint does not begin with guesswork. For several sizes, scale the repaired mesh rather than re-generating, because a second generation returns a different character.
A realistic first attempt is a mesh that looks right on screen and needs thirty minutes to two hours of repair and setup before it prints cleanly, most of that going into thickness and supports. Decide early whether the figure must be a physical solid or only a render; the rest of the pipeline follows.
Yes, as a starting point. An image-to-3D engine can build a plausible mesh from a single view, but everything the camera could not see is invented. Expect to add one or two extra views, then spend time on repair before the file is safe to slice.
Not strictly, but detail is the reason most people choose one. Resin holds features down to roughly 0.4 mm at 0.03-0.05 mm layer heights. FDM with a 0.4 mm nozzle typically needs about 0.8 mm features, which smooths out faces, fingers and thin accessories.
Collector scales commonly run from 1:18 to 1:6. A 1:12 figure is around 150 mm and a 1:6 figure about 300 mm. Both print fine on common resin machines; above roughly 200 mm, splitting the model into parts usually beats printing one tall piece.
Because the engine only saw the front. With one input image, the rear of the head, the far arm and anything occluded are guesses, and two runs of the same prompt can guess differently. Supplying two to four views of the same character as multiview input is the practical fix.
For resin printing, usually yes: hollowing to a 1.5-2.5 mm wall saves material and reduces peel forces, but it needs two or three drain holes of 2-3 mm placed where they will not show. Small solid parts and FDM prints are often left solid instead.
STL or 3MF. STL carries geometry only and is the most widely accepted; 3MF keeps units, colour and multiple objects in one file, and Bambu Studio, OrcaSlicer, Cura and PrusaSlicer all read it. Export after repair and scaling, not before.
Not reliably. Generators return a surface that may contain holes, self-intersections and interior shells, and a slicer that repairs them silently can hide a thin patch that fails later. Treat cleanup as part of the process, and print a small test piece before a long job.
Send the same prompt or image to several engines, compare the meshes side by side, then export GLB, glTF, OBJ, STL, PLY or 3MF. Creating an account grants free credits and needs no card.
A photograph carries no depth, so the printed result depends on how much the software is allowed to guess. This guide covers capture, three routes to a printable solid, the mesh checks that matter, and the scale and wall thickness a small figure needs.
Four routes to a figure mesh, the repair pass a slicer quietly expects, and the resin versus FDM decision, with typical numbers for scale, wall thickness and layer height.
Silhouette extrusion, lithophane height maps and AI image-to-3D all end in the same file extension and almost nothing else. Here is what each route controls, which photos survive it, and the checks that happen before the first layer.