Transform a single photo, product shot, or illustration into an interactive 3D Gaussian Splat (.ply) in seconds. Upload your image, preview in 360°, and download production-ready 3D models online.
1,048,576
Max 3D Gaussians per single image
1.80B
DiT + DINOv3 + FLUX.2 VAE + BiRefNet
0.885 CLIP
Outperforms TRELLIS & TripoSplat
Standard .PLY
Direct y-up 3DGS export for any engine
1.80B Params · DINOv3 + FLUX.2 VAE · 3DGS .PLY
Loading 3D Gaussian Splat... 0%
Decoding spherical harmonics, opacities, and 3D Gaussian covariance matrices
.PLY|Mode: synthetic · Steps: 50 · CFG: 3.0 · Seed: 42Go from a single 2D photograph, product picture, or AI illustration to a full 360° 3D Gaussian Splatting model in three simple steps.
Upload a JPG, JPEG, PNG, or WebP picture of an object. Choose Synthetic mode (1/1.2 crop) for renders and AI art, or Real Photo mode (1/1.7 crop) for camera photos. Built-in BiRefNet segmentation removes the background automatically.
Idol 3D AI encodes your image through DINOv3 ViT-H+/16 and the FLUX.2 VAE, runs 50 Euler flow steps on its 370.6M-parameter DiT denoiser, and decodes octree anchors into up to 1,048,576 3D Gaussians—completing the back and sides your photo never showed.
Rotate, zoom, and compare the 2D source image against the live 3D Gaussian Splat inside our interactive WebGL viewer. Export the standard .ply file for Blender, Unreal Engine, Unity, or web 3D viewers.
Explore 36 single-image to 3D reconstructions across Objects, Furniture, Food, Toys, and Digital Art. Click any card to inspect the interactive 3D Gaussian Splat or download its .ply file.
Generated from a single front-three-quarter studio render into 262,144 3D Gaussians. Idol 3D reconstructs the hidden back straps, stitching, side pockets, and brass buckles in full 360° volumetric detail.
Complex metallic reflections and fine optical cylinders reconstructed via DINOv3 ViT-H+/16 and FLUX.2 VAE conditioning. Preserves thin structural arms and mirror assembly without mesh tearing.
Multi-part footwear asset featuring urethane wheels, laces, eyelet hooks, and ankle boot contour synthesized from a single 2D image.
Demonstrates Gaussian Splatting superiority over traditional polygon meshes on translucent glass, specular highlights, and ornate atomizer hardware.
E-commerce appliance visualization with glossy enamel housing and brushed stainless steel bowl synthesized in 50 Euler flow steps.
Sharp geometric edges, optical lens housing, and layered instant prints reconstructed from a single perspective.
Official Idol 3D showcase piece. Captures plush upholstery folds, curved wooden armrests, and unseen rear backrest geometry with 262,144 anisotropic 3D Gaussians.
Deep button-tufted leather grain, rolled arms, and warm patina reconstructed for interior design and spatial commerce previews.
Continuous organic shell curvature and four-star aluminum swivel base synthesized with accurate concavity and cushion depth.
Unlike conventional polygon mesh generators that produce over-smoothed geometry and muddied UV maps, Idol 3D AI directly synthesizes dense 3D Gaussian primitives with sharp multi-view fidelity.
Scale your output resolution from 65,536 splats for instant web previews up to 1,048,576 Gaussians for intricate surface textures and fine geometry.
Trained as a latent sequence multimodal flow model, Idol 3D AI reconstructs plausible rear geometry, straps, handles, and back surfaces never visible in the input photo.
Combines semantic 3D structure from Meta's 840.6M DINOv3 ViT-H+/16 encoder with fine pixel-level color and texture detail from the 84.1M FLUX.2 VAE.
No manual masking needed. A dedicated 220.7M BiRefNet segmentation network isolates your subject and applies optimal 1/1.2 or 1/1.7 framing automatically.
Exports universal binary_little_endian .ply files rotated to the y-up convention, with clamped opacities and log-scales for seamless third-party import.
A 50.8M octree probability decoder places spatial anchors across active surface voxels, and a 237.3M Gaussian decoder expands each anchor into 32 anisotropic splats.
Powered by our Idol 3D 1.80-billion-parameter architecture, our pipeline couples flow-matching diffusion with sparse octree anchor decoding for rapid, high-precision 3D synthesis.
In multi-view evaluations across 1×H100 GPU benchmarks, Idol 3D achieves the highest structural similarity (0.726 SSIM), lowest perceptual error (0.213 LPIPS), best silhouette overlap (0.862 IoU), and strongest novel-view semantic alignment (0.885 CLIP and 0.781 DINOv2)—outperforming TripoSplat, Microsoft TRELLIS, and LGM.
Input view (SSIM, LPIPS, IoU) & 7 novel views (DINOv2, CLIP, Sharpness)
| Model | SSIM ↑ | LPIPS ↓ | IoU ↑ | DINOv2 ↑ | CLIP ↑ | Time ↓ |
|---|---|---|---|---|---|---|
| Idol 3D50 steps · 262,144 splats | 0.726 | 0.213 | 0.862 | 0.781 | 0.885 | 6.8 s |
| TripoSplat50 steps · 262,144 splats | 0.712 | 0.225 | 0.860 | 0.743 | 0.868 | 9.9 s |
| TRELLISPaper setting · 50+50 steps | 0.716 | 0.235 | 0.845 | 0.754 | 0.868 | 7.1 s |
| LGMDefault · 65,536 splats | 0.722 | 0.237 | 0.834 | 0.575 | 0.784 | 1.1 s |
Whether you are prototyping game environments, showcasing products in 360°, or turning 2D concept art into spatial assets, Idol 3D AI accelerates every stage of 3D content creation.
Convert 2D prop sheets, isometric buildings, and character concepts into 3D Gaussian Splats for rapid level blockouts and background assets in Unreal & Unity.
Transform standard catalog photos of footwear, handbags, furniture, and electronics into interactive 360° web previews that boost customer engagement.
Bring Midjourney, FLUX, or hand-drawn illustrations into three dimensions to check lighting, camera angles, and spatial proportions in seconds.
Export standard .ply splat clouds for WebXR, Apple Vision Pro, and Gaussian Splatting plugins without spending hours on multi-camera photogrammetry.
Idol 3D AI is an online AI Image to 3D Generator powered by a 1.80-billion-parameter 3D foundation model. It transforms a single 2D photograph, product image, or digital illustration into up to 1,048,576 3D Gaussians. The pipeline automatically removes the background with BiRefNet, extracts deep semantic and spatial features using DINOv3 ViT-H+/16 and the FLUX.2 VAE, samples a 370.6M-parameter Diffusion Transformer (DiT), and decodes an octree of anchors into a standard 3D Gaussian Splatting (.ply) model—including the back and sides the original image never showed.
Idol 3D AI outputs industry-standard 3D Gaussian Splatting (.ply) binary files containing 3D positions, normals, degree-0 spherical harmonics color (f_dc_0, f_dc_1, f_dc_2), logit opacity, log scale, and unit quaternion rotation in a y-up coordinate system. You can inspect the .ply model right in your browser or import it into Blender, Unreal Engine, Unity, SuperSplat, Luma, or Polycam.
The input mode controls how tightly the preprocessor frames your subject after background removal. Synthetic mode (1/1.2 crop ratio) is optimized for 3D renders, digital illustrations, and AI-generated images of isolated objects. Real Photo mode (1/1.7 crop ratio) leaves additional spatial margin around real-world smartphone or camera photographs to account for natural lens perspective.
Traditional 2D-to-3D mesh generators force neural reconstructions through Marching Cubes or Poisson surface extraction, which often blurs fine textures, translucent surfaces, fur, or thin geometry. 3D Gaussian Splatting represents the object directly with hundreds of thousands of anisotropic 3D Gaussians, preserving crisp multi-view appearance, realistic lighting gradients, and sharp silhouettes at 60 FPS.
Idol 3D AI supports between 32,768 and 1,048,576 Gaussians. The default 262,144 Gaussians offers the best balance of sharp 360-degree surface detail and fast download size (~17 MB). For lightweight web embeds, 65,536 or 131,072 Gaussians load almost instantly, while 524,288 and 1,048,576 Gaussians are ideal for hero assets and close-up inspection.
Yes. All .ply 3D models generated on our platform come with full commercial usage rights and can be integrated into commercial games, e-commerce storefronts, marketing visuals, digital art, and AR/VR spatial computing applications.
Upload any photo or illustration and experience real-time 3D Gaussian Splatting with Idol 3D AI—right in your browser.