Skip to content

How to prepare an image for image to 3D

The mesh an image-to-3D model returns can only be as clean as the picture it reads. Four checks before upload fix most bad results: one subject, the whole subject in frame, a plain background and soft, even light.

Four checks to prepare an image for image to 3D

Each check you make to prepare an image for image to 3D removes one thing the model would otherwise have to guess. In each pair, avoid the picture on the left; the one on the right is the actual input behind a model generated in AI3D Studio.

Left: the tin robot on a crowded toy shelf, gripped by a hand. Right: the same robot alone on a plain gray backdrop.

Crop to one subject

Left: a robot toy on a crowded toy shelf, held by a hand · Right: the same robot alone

Remove other objects, hands holding the subject and busy props. Anything left in the picture may be fused into the mesh.

Left: a white sneaker framed so tightly that the toe and heel are cut off. Right: the whole sneaker with space on every side.

Keep the whole subject in frame

Left: framed so tight the toe and heel are cut off · Right: the whole shoe with a margin

Leave a margin on every side. A cut-off foot or handle comes back as a flat stump or an invented shape.

Left: a succulent in a terracotta pot among patterned mugs and a flowered curtain. Right: the same pot on a plain gray backdrop.

Use a plain background

Left: a flowered curtain and mugs behind the pot · Right: one plain gray backdrop

A single color that contrasts with the subject makes the outline unambiguous, and the outline sets the silhouette of the model.

Left: a dragon statue under a single hard spotlight with deep black shadows. Right: the same statue in soft, even light.

Light it softly and evenly

Left: one hard spotlight and deep shadows · Right: soft, even light

The texture is painted from the pixels, so hard shadows and bright highlights can end up baked into the surface color.

What six prepared images became in 3D

The four right-hand pictures from the checklist above, the three-quarter chest from the next section and one clean character render, each run once through image to 3D in AI3D Studio. The inset is the prepared input; open a card to orbit the model and judge the outline, the painted texture and the sides the camera never saw.

Uploaded picture: Tin robot

Tin robot

Tripo H3.1

Input: the robot alone on plain gray, with the shelf and the hand cropped away. Tripo H3.1 reads JPEG or PNG up to 20 MB.

Uploaded picture: Running sneaker

Running sneaker

Hunyuan3D 3.1 Pro

Input: the whole shoe with a margin on every side, so the toe and heel are in the picture. Hunyuan3D 3.1 Pro takes JPEG, PNG or WebP up to 8 MB.

Uploaded picture: Succulent in a pot

Succulent in a pot

TRELLIS.2

Input: the pot on one plain backdrop, with the curtain and mugs gone. TRELLIS.2 takes JPEG, PNG or WebP up to 20 MB.

Uploaded picture: Dragon statue

Dragon statue

Hitem3D 3.0

Input: the statue in soft, even light instead of one hard spotlight. Hitem3D 3.0 takes images up to 20 MB and can export STL for printing.

Uploaded picture: Treasure chest

Treasure chest

Hunyuan3D 3.1 Rapid

Input: a three-quarter view that shows the front, one side and the open lid, so less of the box is inferred. Hunyuan3D 3.1 Rapid takes JPEG, PNG or WebP up to 8 MB.

Uploaded picture: Chibi knight

Chibi knight

Tripo H3.1

Input: a clean character render with one subject, the whole figure in frame and flat studio light, which passes all four checks as it is.

What the upload field accepts

2 formats everywhere

JPEG and PNG work on every model; WebP works on all except Tripo and Meshy, which refuse it.

8–20 MB per image

20 MB on most models; Hunyuan3D 3.1 Pro, Rapid and Sketch take up to 8 MB.

1 subject

One object or character per picture; a scene is read as one lumpy object.

Up to 8 views

When one picture hides too much, Hunyuan3D 3.1 Pro reads up to eight views of the same subject.

What the upload field accepts, model by model

Check the size cap of the model you plan to use before you export the picture. A model that reads several views takes them as separate pictures of the same subject.

ModelFile typesLargest filePictures per run
Tripo H3.1JPEG, PNG20 MBOne image, or up to 4 views
Tripo P1 and P2JPEG, PNG20 MBOne image
Meshy 7.1JPEG, PNG20 MBOne image, or up to 4 views
Hunyuan3D 3.1 ProJPEG, PNG, WebP8 MBOne image, or up to 8 views
Hunyuan3D 3.1 RapidJPEG, PNG, WebP8 MBOne image
Hunyuan3D SketchJPEG, PNG, WebP8 MBOne drawing plus a description
TRELLIS.2JPEG, PNG, WebP20 MBOne image
Rodin Gen-2.5JPEG, PNG, WebP20 MBOne image, or up to 5 views
Hitem3D 3.0JPEG, PNG, WebP20 MBOne image, or up to 4 views
Seed3D 2.0JPEG, PNG, WebP20 MBOne image
Left: a treasure chest photographed straight-on so only its front face shows. Right: the same chest from a three-quarter view showing the front, side and lid.

Why the checks matter

Why prepare an image for image to 3D: what the model reads

Knowing which part of the picture drives which part of the model tells you what to fix first.

The outline becomes the shape

The silhouette against the background decides the overall form, so a clean outline is the first thing to get right.

The pixels become the texture

Color and lighting in the photo are projected onto the surface. Rodin Gen-2.5 has a remove-lighting option for photos with strong light.

The camera angle sets the pose

A three-quarter view shows more depth than a flat front view. Tripo H3.1 can align the model to the image's orientation.

Hidden sides are inferred

Anything the camera cannot see is filled in by the model; no amount of preparation changes that for a single picture.

What a good input image cannot fix

Taking time to prepare an image for image to 3D removes avoidable errors; it does not turn a single picture into a measurement.

No real size

A photo carries no scale. Tripo H3.1's real-world scale switch estimates size in metres on textured models; otherwise set the size in your 3D tool.

Results still vary

The same prepared image can give different geometry on a second run; a fixed seed repeats a result on models that expose one.

Reflections become paint

Glass, chrome and mirrors show their surroundings instead of their own color, and those reflections are painted onto the surface like any other pixels.

Fine strands merge

Hair, fur, cables and thin handles tend to fuse or break however clean the picture is.

Questions about input images for 3D

Short answers about files, backgrounds and generated pictures.







Turn the prepared image into a model

Upload it on the image to 3D page and compare the credit cost of each model before you submit.