Uploading & Preparing Data for Splatware
In this section, you'll learn everything about uploading your source material to Splatware to create photorealistic and ultra-optimized 3D models. Understanding what to upload, best practices, and how to prepare your files is essential for achieving the best results.
Supported File Formats
Splatware supports a wide range of file formats for both image and video uploads. Models can also be imported from external sources.
| File Type | Formats Supported |
|---|---|
| Images | JPG, PNG, HEIC |
| Videos | MP4, MOV, MKV, FLV, AVI, 3GP |
| 3D Models | PLY, SPLAT (.splat) |
Recommended Minimum Input Data
For best results in model generation, Splatware recommends the following input data volumes:
- Minimum 20+ high-resolution images from multiple angles OR
- Minimum 20+ seconds of video with good coverage
- Good lighting & stable footage highly recommended
- 360 Cameras produce optimal results
The more diverse your data angles and perspectives, the better the final 3D output.
Uploading & Data Preparation
Splatware provides a seamless and powerful workflow for uploading imagery and generating high-fidelity 3D Gaussian Splatting models. Whether you're a creator or a studio, the process at /create is streamlined for speed, quality, and control.
Upload Steps
- Go to
/create - Drop your images into the uploader (200–1.000 recommended)
- Select your AI Model type
- Click "Create Project" to launch training
The system performs automatic Structure-from-Motion, segmentation, and optimization, exporting results as lightweight .SPLAT files optimized for real-time rendering.
AI Model Types
General Model (Free)
Fast and accessible. Ideal for general scenes and casual creators.
- Training time: ~5 minutes
- High resolution, minimal compression
- File size: 6–8MB
- Ultra-fast AI training pipeline
Cinematic Model (Premium)
Ultra high resolution. Designed for film, product, and creative studios.
- Training time: ~30 minutes
- Highly detailed splats with depth fidelity
- Precise camera mapping
- Larger file output for highest quality
Segmentation Model (Premium)
Focuses on isolated object reconstruction with AI-powered background removal.
- Training time: ~20 minutes
- AI auto object detection and masking
- Manual refinement supported
- Best for clean, closed objects (e.g. tools, statues)
- Complex objects (e.g. foliage, plants) may lose detail
Large Model (Enterprise)
Engineered for massive indoor/outdoor reconstructions and enterprise needs.
- Training time: 1+ hour
- Super-dense splats with maximum spatial coverage
- Custom optimization layers
- Finetuning options available
- Ideal for architecture, urban scans, and research
Output Format
Every training result is exported in .SPLAT format — an efficient, GPU-native representation that enables real-time rendering with low latency, even for large scenes.
Supported Data Sources
You can upload data captured from:
- Mobile Phones (any camera)
- 360 Cameras (Insta360, GoPro, Ricoh etc.)
- Drones
- DSLR / Professional Cameras
- Existing 3D Scans (PLY or SPLAT)
All uploaded content gets converted for optimal online performance across Web, Mobile, VR & AR platforms.