How to prepare an image for 3D generation
Choose a useful angle, simplify the background and understand what one photo cannot tell an AI model.
A single photograph is a starting point for a 3D shape, not a measurement of it. Hunyuan 3D estimates the surfaces that are not visible. Better inputs reduce ambiguity, but they do not turn a photograph into an exact scan.
Choose one subject
Use an image with one clearly separated object. A group of objects, a collage or a busy room makes it harder to determine what belongs to the model. If you want a small figurine, crop around that figurine rather than including the shelf it sits on.
Leave a little space around the silhouette. Do not crop off ears, handles, feet or other important protrusions. Those details help define the shape.
Show more than a flat front
A three-quarter view usually communicates depth better than a straight-on picture. It can reveal the front, a side and part of the top at once. This is a practical starting point, not a promise that every object will reconstruct better from the same angle.
For a transparent glass, a shiny metal part or a thin wire frame, a single photograph can be especially ambiguous. Reflections and shadows may be interpreted as geometry. Try even lighting and a neutral background, and expect to inspect the result carefully.
Use a supported image
The Image to STL workbench accepts JPG, PNG and WebP files up to 6 MB. Each side must be between 128 and 5000 pixels. A larger image does not automatically produce a more accurate mesh; sharp edges, visible shape and a clear subject are more useful than extra background pixels.
Selecting a file only creates a local preview. Starting a live generation sends the image to our server and Tencent Cloud. Use images you have the right to process, and avoid sensitive material.
Expect hidden surfaces to be invented
An AI cannot recover the unseen back of an object from direct evidence in a single image. A handle might attach differently, an opening may be filled in, or a small feature may disappear. Rotate the generated model before deciding to export it.
For functional parts, use actual measurements and CAD. A generated mesh is not a substitute for a dimensioned drawing, a calibrated scan or a tolerance check.
Inspect before printing
After generation, use the solid view for the overall silhouette and wireframe for the mesh structure. Set a longest-side dimension in millimeters, then export STL. Open the file in your slicer and check for disconnected parts, holes, thin walls and required supports.
Changing the longest side scales all axes uniformly. It does not correct the proportions that the model inferred from the image.
Open the Image to STL workbench, or read the mesh checks to make before printing.
Match the reference to its destination
For a game prop, preserve the silhouette that matters at gameplay distance. For an ecommerce model, keep reference photos and measurements for all visible product features. For a website viewer, plan to retain a suitable GLB and its materials.
If your goal is a flat badge or a backlit portrait, the input preparation is different: read AI, heightmaps, logos and lithophanes before choosing a converter. For the broader process, start with the image-to-3D guide.
Sources
Technical background checked November 18, 2026: Tencent Hunyuan3D-2 documentation. File limits and workbench behavior above describe Image to STL's current implementation, which uses a separate hosted integration.