Overview
Meshy's Image to 3D feature converts a single reference image into a detailed 3D model. The quality of your output depends heavily on the quality of your input. This guide walks you through the key factors that give you cleaner geometry, better textures, and more accurate results.
Using Meshy 6 or Meshy 7? Geometry quality has improved significantly — especially for characters (smoother anatomy) and hard surfaces (sharper edges). Many issues that required workarounds in earlier versions may already be resolved.
Meshy Image to 3D: Better Results Checklist
Use this checklist before every generation to maximize output quality.
✅ Image Quality
Use a high-resolution image (at least 512 × 512 px; 1024 × 1024 px or higher is ideal)
Ensure the subject is sharp and in focus — blurry images produce noisy geometry
Avoid heavy JPEG compression artifacts
✅ Background
Use a plain white or transparent background for the cleanest object isolation
If your background is busy, remove it first using a tool like remove.bg before uploading
Avoid reflective or gradient backgrounds — they can bleed into the model's texture
✅ Subject Framing & Angle
Frame the object so it fills most of the image — leave minimal empty space around the edges
Use a front-facing or slightly angled view for the best 3D reconstruction
Avoid extreme top-down or bottom-up angles — these limit the model's ability to infer depth
For characters, a T-pose or neutral standing pose produces more usable topology
✅ Lighting
Use even, diffused lighting with no harsh shadows
Avoid strong directional light — shadows bake into the texture and make it look flat in other scenes
Studio-style or soft-box lighting gives the best results
✅ Subject Complexity
Start with simple, clearly defined objects if you're new to Image to 3D
Highly transparent or reflective objects (glass, mirrors) are difficult to reconstruct — use opaque references instead
Fine details like hair strands or thin wires may not fully reconstruct — for highly detailed organic shapes, Text to 3D often gives better control
Recommended Image Settings
Setting | Recommended Value | Why It Matters |
Resolution | 1024 × 1024 px or higher | More pixel data = sharper texture |
Background | White or transparent (PNG) | Cleaner object isolation |
File format | PNG (preferred) or high-quality JPG | Avoids compression artifacts |
Subject fill | 70–90% of frame | Maximizes usable detail |
Lighting | Even, diffused | Prevents shadow bake-in on texture |
Using Multi-View Input
Multi-view input is one of the most effective ways to get clean geometry on all sides of your model. When Meshy only sees one angle, it has to infer what the unseen sides look like — which can produce "hallucinated" geometry that doesn't match your intent.
When to use it: Any time your subject has significant detail on multiple sides — characters, vehicles, props, creatures, or anything where the back matters as much as the front.
Ideal angles to capture:
Front view
Side view (left or right)
Back view
¾ angle (optional, but helpful for complex shapes)
Tips: Keep all reference images consistent in style, lighting, and scale. Mixing photos from different sessions or scales can confuse the reconstruction and produce mismatched seams.
Using Prompt Refinements
After uploading your image, you can add a text prompt to guide the generation. Use prompts to:
Specify a style: "photorealistic surface", "cartoon character"
Request low-poly output: enable Low Poly Mode in the generation settings for game-ready assets with clean, optimized geometry — ideal for real-time use in engines like Unity or Unreal
Describe unseen sides: "the back of the character has a cape"
Clarify material: "metallic armor", "worn leather texture"
Keep prompts concise and descriptive. Overly long prompts can dilute the influence of your reference image.
Tips for AI-Generated Reference Images
Many users generate their reference image with tools like Midjourney or DALL-E before uploading to Meshy. If that's your workflow, add these to your image generation prompt to get a Meshy-ready reference in one shot:
White background — skips the background removal step entirely
Neutral, even lighting — avoids shadow bake-in on the final 3D texture
Front-facing pose — gives Meshy the clearest view of the subject's full silhouette
Example prompt addition: "white background, studio lighting, front view, product photography style"
Troubleshooting
The model looks lumpy or has poor geometry
This usually means the input image had a busy background or low contrast between the subject and surroundings. Remove the background and re-upload with a clean white or transparent background. If you're on Meshy 6 or Meshy 7, try regenerating — geometry quality has improved significantly and may produce a better result without any changes.
Textures look flat or washed out
Strong directional shadows in your reference image bake into the texture. Switch to a more evenly lit reference photo, ideally shot under diffused or studio lighting.
Missing geometry on one side
Meshy infers unseen geometry from context. If one side is heavily obscured, add a text prompt describing what should appear there. For best results, use multi-view input with front, side, and back angles — this gives the model real reference data instead of guesses.
The object is cropped or clipped
Reframe your image so the full subject is visible with a small margin around all edges. No part of the object should touch or extend beyond the image boundary.
Frequently Asked Questions
Can I use a photo taken on my phone?
Yes — phone photos work great as long as the subject is well-lit, in focus, and photographed against a clean background. Use portrait or studio mode if available to reduce background clutter.What is the minimum image resolution?
At least 512 × 512 px. For best results, use 1024 × 1024 px or higher to give the AI more detail to work with.Does my image need a white or transparent background?
It's strongly recommended. A plain background helps Meshy isolate the subject correctly and prevents background textures from bleeding into your model.Why does my model have holes or missing geometry?
This usually happens when part of the subject is cut off, heavily shadowed, or obscured. Reframe the image so the full object is visible, or add a text prompt describing the missing areas. Using multi-view input (front, side, and back) is the most reliable fix.When should I use Text to 3D instead of Image to 3D?
Image to 3D is best when you have a clear reference and want to match a specific look. Switch to Text to 3D when your subject has fine details that don't photograph well (hair strands, thin wires, fur), when you want full creative control over unseen sides, or when you don't have a suitable reference image and want to describe the object from scratch.Can I upload multiple reference images?
Yes — multi-view input is supported on select plans and is highly recommended for complex subjects. Upload front, side, and back angles to give Meshy real geometry data on all sides, reducing hallucinated or mismatched geometry.How long does Image to 3D generation take?
Most generations complete in under 2 minutes. Complex or high-detail subjects may take slightly longer depending on server load.What export formats are available?
You can download your completed model in GLB, FBX, OBJ, USDZ, STL, BLEND, and 3MF formats to suit any workflow.Transparent or glass objects don't look right — what can I do?
Transparent and highly reflective surfaces are difficult for AI to reconstruct accurately. Use an opaque reference of the same shape, or switch to Text to 3D with a detailed material description instead.