Preparing Images for Image-to-3D: Background, Angle, Lighting

Meshy’s documented rules

Meshy’s Multi-View Image to 3D tutorial (accessed 2026-11-08) says:

• Upload one to four photos of the same object from different angles: a main image plus Left, Back and Right slots. Recommended angles are front, side, back and a three-quarter angle.
• A single photo works, but two to four angles “noticeably improve results” according to Meshy.
• Use a plain background (white, gray or transparent), even and diffused lighting, the object centered and filling the frame, and the same camera distance for each shot.
• Avoid backgrounds close to the object’s color, strong highlights, smoke or fog, and watermarks or captions.
• Minimum 1040 x 1040 pixels, PNG, JPG, JPEG or WebP, up to 20 MB each, one object per photo and one angle per slot. A single reference sheet showing front, side and back can be uploaded as the main image.
• Multi-View needs a Pro plan or above, runs only on Meshy 7, and is unavailable in Smart Topology mode.

These are Meshy’s instructions for Meshy. The “improve results” statement is a vendor claim with no data on the page.

Meshy’s API limits for the same input

The Image to 3D API (accessed 2026-11-08) states hard limits that differ from the web tutorial’s advice, so keep them apart:

• image_url is a public URL or a base64 data URI such as data:image/jpeg;base64,....
• Formats: JPG, JPEG, PNG or WebP. At most 20 MB, at least 32 px on each side, and at most 178,956,970 pixels (width x height).
• The API page says nothing about background handling; it mentions moderation only. With moderation: true, a flagged input returns 400, and a 403 means moderation blocked an input image.
• image_enhancement (default true; works only with meshy-6, meshy-7.1 or latest; 400 with meshy-6-lite or Smart Topology). The page says false keeps the exact look of the input image, with no style processing. If your reference is finished concept art you do not want restyled, this is the documented switch.
• remove_lighting (default true) removes lighting from the base color and takes effect only with the Meshy 6 texture model.
• should_texture defaults to true; false returns an untextured mesh.

So there are two floors: 32 px is what the API accepts, 1040 x 1040 is what Meshy’s tutorial recommends. Which produces better meshes is not documented.

Other tools: what is documented

Rodin’s Gen-2.5 API page adds that you submit one to five images, the first image is used for material generation, and the prompt is optional (the service derives guidance from the images); it does not state accepted formats, file size, or resolution limits. Tripo’s developer page lists Image to 3D (20 to 30 credits) and Image to Multiview (10 credits) only as price items and gives no input requirements. Sources are the vendors’ docs and READMEs, dated 2026-11-08. “None found” means we did not see it, not that none exists.

What one vendor guide adds

3D AI Studio’s guide on single versus multiple photos is undated and vendor-authored. It says single-photo generation is faster but less accurate, recommends three to eight photos with six to eight as the best balance, and says sharp, well-lit images matter regardless of count: “One excellent high-res photo can beat three blurry low-res photos.” It does not address backgrounds. It also quotes accuracy percentages with no stated method, which we do not repeat. Its photo counts differ from Meshy’s one-to-four limit, which is a reminder that the right number depends on the tool.

A preparation checklist from the documented rules

1. Read your tool’s own limits first (count, size, formats).
2. One object per image, centered, filling the frame.
3. Plain background that does not match the object’s color, per Meshy.
4. Even lighting, no hard highlights, no fog, no text overlays.
5. For multi-view, same distance and same lighting for every angle.
6. Use the resolution floor your tool states, and do not upscale a small photo to pass it; the sources do not say whether that helps.
7. If the input is concept art, a front, side and back sheet is the multi-view equivalent; see concept art to a 3D character.

Troubleshooting

Causes come from the vendors’ stated rules and general reasoning; none was reproduced.

What is not documented

We found no source that quantifies how much each step changes the mesh. Nothing we read covers sketches in detail, transparent versus white backgrounds, or shadows. Hidden-surface behavior, such as what a generator invents for the back of an object from one photo, is described by vendors but not shown. The test pages we plan, multi-view versus single image and photo and concept-art input, have not been run.

To pick a tool for image input, see the image-to-3D shortlist. For text input instead, read writing prompts.

See also the AI 3D generation guides hub.