When an AI 3D model comes out lumpy, with fused arms or a melted back, the model is rarely the problem. The input image is. After a lot of image-to-3D runs (locally with TRELLIS.2 and on JollyAI), these are the rules that make the biggest difference.
Disclosure: JollyAI is my own tool. These tips work for any image-to-3D model.
The 9 Rules
1. One subject only
Two characters or a character holding a big prop become one fused blob. Crop or generate a single subject.
2. Show the whole thing
Cut-off feet, hands or the top of the head means the model has to invent them, and it usually does badly. Leave a margin around the subject.
3. Plain background
Solid white, grey or a removed background. Busy scenes leak into the mesh. See my background removal guide.
4. Arms away from the body
Arms touching the torso get merged into it, and then they can’t be rigged. An A-pose (arms about 30 to 45 degrees down) is ideal.
5. Front or three-quarter view
Straight side views and extreme top-down angles hide too much. A slight three-quarter angle helps props show their depth.
6. Even, soft light
Hard shadows get baked into the texture and can be read as shape. Soft studio light is best.
7. Smooth style, not pixel art
Illustrations, 3D renders, anime and photos work. Pixel art and voxel art don’t carry enough shading to guess depth. Want a blocky result? Use a smooth input and the Blocky / voxel option.
8. Simple, readable shapes
Thin straps, lace, chains, see-through fabric and flyaway hair come out blurry or merged. Chunky shapes and clear silhouettes win.
9. Sharp and big enough
Use at least about 1024 px on the long side, in focus, without heavy compression or motion blur.
Prompts That Generate Perfect Inputs
| Asset | Prompt |
|---|---|
| Character | full body [character], standing in A-pose, arms slightly away from body, open hands, front view, plain light grey background, soft studio lighting, clean 3D render |
| Anime avatar | full body anime [character], front view, A-pose, simple outfit, plain white background, clean cel shading, sharp lines |
| Creature | full body cute [creature] standing on two legs, front view, arms out, plain background, soft light, stylised 3D render |
| Prop | single [object], three-quarter view, centred, plain white background, soft studio lighting, product render |
You can generate these with Krea 2 or Qwen 2.1 on JollyAI, then send them straight to Image to 3D on the same plan.
Pre-Flight Checklist
- ☑ One subject, fully in frame
- ☑ Plain background
- ☑ Arms away from the body (A-pose if you’ll rig it)
- ☑ Front or three-quarter view
- ☑ Soft, even light
- ☑ Smooth style, simple shapes
- ☑ At least about 1024 px, sharp
Got a clean image? Test it now. 1 free 3D preview a day, with optional Animate it and Blocky / voxel.
Try Image to 3D free →
Common Problems and the Input Fix
| Problem in the 3D model | Fix the image by |
|---|---|
| Arms fused to body | Regenerate in A-pose |
| Background chunks stuck to the model | Removing the background |
| Flat, thin model | Using a shaded, smooth style instead of flat or pixel art |
| Weird back side | Simpler outfit and hair; three-quarter view |
| Blurry texture | Higher resolution, sharper input |
| Shadows painted on the texture | Softer, even lighting |
What You Get With PRO (and the Cheaper Options)
| Free | Day Pass | PRO | |
|---|---|---|---|
| Price | $0 | $3.99 for 24 hours | $14.99/month for the first 50 members, locked for 12 months (then $19.99) |
| Generations | 1 per day | 25 | 50 per day |
| AI video (MiniMax H3: text, image and reference to video, with sound) | – | Up to 3 | Up to 10 per day |
| AI images (Krea 2, Qwen 2.1) and image editing | – | Yes | Yes |
| Image to 3D (with Animate it and Blocky / voxel) | Watermarked 360° preview | GLB download | GLB download |
| 3D Rigger (body, fingers, face, hair) | Free preview | FBX + GLB download | FBX + GLB download, 10 rigs a day |
| Text to voice | Generate and listen | Generate and listen | Generate and download |
Why the bundle is good value: you generate the perfect input image, edit it, turn it into 3D and rig it on one $14.99 plan, instead of juggling an image generator, an editor, a 3D tool and a rigger.
Final Thoughts
Spend two minutes on the input and you’ll save ten re-generations. Follow the checklist, and most image-to-3D models will come out usable on the first try.
Put these tips to work. Free daily preview, no GPU.
Open JollyAI Image to 3D →
JollyAI is my own tool. Download with a Day Pass ($3.99) or PRO ($14.99/mo for the first 50 members).