JPG to 3D Model: Accepted Image Files | TRELLIS.2

Does a JPG Work as an Input?

Yes. A JPEG is one of the accepted image types for an image-to-3D generation on this service. The picker takes image/jpeg, image/png, image/webp, image/gif, and image/avif, up to 10 MB per file. Anything outside that list is rejected before a job starts, so nothing is charged for a file the service cannot read.

That answer is short because the interesting part is not the extension. It is what JPEG compression and the phone that produced the file do to the pixels the model gets to see.

What JPEG Costs You Before the Model Ever Sees It

JPEG is a lossy format. It throws away detail, especially around hard edges, and the loss multiplies when a picture is saved and re-saved between apps.

For a generation that infers geometry from a silhouette, the damage is specific:

• Ringing and blocking around the object boundary can blur the outline the model relies on
• Fine texture, thin lines, and small details soften before the job begins
• A screenshot of a JPEG preview is a second compression pass on an already compressed image
• A "save for web" export at low quality is a third

PNG or WebP avoid most of that, and a PNG can also carry transparency, which keeps the silhouette unambiguous. JPEG is not disqualifying; a good JPEG beats a bad PNG.

Phone Photos: HEIC, Size, and Orientation

• HEIC. iPhone cameras write HEIC. iOS Safari usually hands the picker a JPEG instead, and a HEIC file that arrives by drag-and-drop is transcoded in the browser before upload. If that conversion fails, the picker says so; the fix is to export the photo as a JPEG and try again.
• Size. The picker accepts up to 10 MB, and files over roughly 4 MB are downscaled before upload. A large photo is therefore resized rather than refused, but you lose control of the crop and the resampling. Cropping the subject yourself gives a better result than letting a full-size frame be shrunk.
• Orientation. Check that the preview appears upright before uploading. A rotation stored only in metadata is easy to lose as a file travels between tools, so export the image the way you want it to look.

Step by Step: JPG to a 3D Model

1. Crop to a single subject, centered, with the whole silhouette inside the frame.
2. Export at a sensible size. Big enough to read the details that matter, small enough that it stays under the downscale threshold if you care about the exact pixels sent.
3. Prefer one clean pass from your editor over a chain of re-saves.
4. Open the image-to-3D generator, sign in, and upload the file.
5. Start at 512³. The credit cost of the selected resolution is shown before you submit, and the task uses available credits without a new first-task allowance.
6. Wait for the tracked job. A standard generation often finishes in about one to three minutes, and failed or canceled jobs return their reserved credits.
7. Inspect the mesh in the GLB viewer, then download the GLB.
8. Convert with the 3D converter if the destination tools want glTF, OBJ, STL, PLY, or VOX.

Common Mistakes

1. Uploading a screenshot of the photo instead of the photo.
2. Re-saving the same file through three apps and losing edge detail each time.
3. Sending a 12 MP frame where the object occupies a fifth of the picture, then letting downscaling throw away the rest.
4. Uploading a sideways photo and blaming the mesh for the orientation.
5. Expecting a cropped composition to be fixed by the model. It only sees what the pixels show.
6. Assuming a bigger file is always a better input. Sharpness and a clean silhouette matter more than megabytes.

Frequently Asked Questions

Can I generate a 3D model from a JPG?

Yes. JPEG is an accepted input type. The generator returns a textured GLB draft from that one image.

Is PNG better than JPG for image-to-3D?

Usually, yes. PNG is lossless and can carry transparency, so the object boundary reaches the model intact. A clean, high-quality JPEG still works well.

How large can the upload be?

The picker accepts files up to 10 MB, and files larger than roughly 4 MB are downscaled before upload. Cropping first gives you more control over what reaches the model.

Can I upload a HEIC photo from an iPhone?

Usually the picker receives a JPEG from iOS automatically, and HEIC files that arrive by drag-and-drop are transcoded in the browser. If the conversion fails, export the photo as a JPEG and upload that.

Does the model use EXIF data?

No. Only the visible pixels matter, which is why the image should be exported upright rather than relying on rotation metadata.

Start from One Clean File

Crop the subject, export it once at a sensible size, and run the cheapest level before spending more credits.

Open the Image-to-3D Generator

Review the full image-to-3D workflow