From PNG
PNG is a lossless format with a true alpha channel, which makes it the best source image for cut-out product shots and objects photographed on a plain or removed background.
Upload a PNG with a clean alpha edge and Hyper3D reconstructs the object in 3D as a textured GLB, which one Blender export turns into the watertight STL your slicer wants.
画像をアップロードして3Dモデルを生成
Cloudflare Turnstileで保護されています。アカウント不要、1時間に3回まで無料で生成できます。
JPG・PNG・WebP対応 · 1時間に3回まで無料 · アカウント不要
ご注意:このジェネレーターが出力するのは、テクスチャ付きのGLBメッシュだけです。そのGLBからSTLにたどり着く手順は、このページで解説しています。
PNG compresses losslessly and, unlike JPG, carries a real alpha channel, so a cleanly cut-out product shot or object photo isolates from its background with none of the fuzzy edge artifacts a lossy format can introduce. Hyper3D uses that clean silhouette as a strong cue, then its AI reconstruction model builds out the full 3D shape of the object, estimating the geometry of the back and sides that a single photo could never show directly.
STL throws away everything but geometry: no color, no texture, no material, just a list of triangles describing a solid surface, which is precisely the input format that FDM and resin printers and slicers like Cura, PrusaSlicer and Bambu Studio expect. Hyper3D returns the textured mesh it reconstructed from your PNG as a GLB, and exporting that to STL strips it down to a closed, watertight triangle shell, since the reconstruction already targets a single manifold surface so the slicer does not choke on gaps.
PNG is a lossless format with a true alpha channel, which makes it the best source image for cut-out product shots and objects photographed on a plain or removed background.
STL discards color and texture entirely and stores only a triangle mesh, a format built in the 1980s for stereolithography that every modern desktop slicer still reads natively.
Use a PNG with the subject isolated on a transparent or plain background so the AI has a crisp outline to build the printable shape from.
The model reconstructs a complete, closed 3D surface from the single view in about 60 seconds, inferring the parts the photo never showed.
Download the GLB, then export it to STL in Blender to get a manifold file stripped of color and texture, ready for Cura, PrusaSlicer or Bambu Studio.
Load the STL into your slicer's repair tool or Meshmixer before printing, since even a clean alpha cutout can leave a thin seam where the reconstruction meets its own silhouette.
Anything steeper than roughly 45 degrees will need support material in PrusaSlicer or Cura, so rotate the model to reduce how much overhang faces downward.
Small features like straps or fingers reconstructed from a PNG can print below your nozzle's minimum width, so scale them up before slicing in Bambu Studio.
AI reconstruction has no absolute sense of real-world size, so always set the model's dimensions manually in your slicer before printing.
The alpha channel does most of the heavy lifting here, so spend the extra minute making sure the cutout edge is genuinely clean before uploading. A rushed auto-background-remover can leave soft, semi-transparent fringe pixels around hair, fur or fabric edges, and those get interpreted by the reconstruction model as part of the object's boundary, which can show up as a thin, unwanted lip on the finished mesh. Re-check the cutout at full zoom in Photoshop or Photos on macOS, and redo the mask by hand around any fuzzy edges before you upload.
After download, bring the STL into Meshmixer to inspect wall thickness and confirm the shell is a single manifold body, since Meshmixer's inspector will flag and can auto-repair the kind of small seam a clean-but-imperfect cutout can leave behind. From there, import into Cura, PrusaSlicer or Bambu Studio, set your layer height and infill, add supports where the AI-reconstructed geometry overhangs, and consider a slight uniform scale-up if delicate parts look thinner than your nozzle can reliably print.
写真を3Dモデルに変換する前によく寄せられる質問です。
It helps a lot. A clean alpha channel gives the AI a precise outline to reconstruct from, though a PNG on a plain solid background with good lighting can also work well.
Yes, Hyper3D includes 3 free generations per hour with no account, no sign-up and no credit card, and Blender, which turns the downloaded GLB into a printable STL, is free as well.
No, STL is a geometry-only format with no material data, so the color and texture in the GLB Hyper3D gives you are dropped the moment you export that file as STL.
Any part of the reconstructed geometry that overhangs by more than roughly 45 degrees will sag or fail without support material, which is normal for organic shapes reconstructed from a photo.
Yes, scale it in your slicer to whatever real-world size you need, since the AI reconstruction has no fixed sense of physical scale and the mesh is just proportional geometry.
Leftover background pixels around the edges can get read as part of the object, so it is worth re-checking the alpha mask at full zoom and cleaning up stray fringe before you upload.
Hyper3Dで画像を3Dモデルにする他の方法。