Speridlabs open-sources Idol: single image to full 3D Gaussian splat in under 2s
pbaylies · x · 2026-11-02
Speridlabs open-sourced Idol, which turns a single image (render, generated image, or photo) into a complete 3D Gaussian splat, automatically removing the background and hallucinating the sides the picture never showed. Key specs: up to 1M Gaussians (1,048,576), 0.66B params, 7.2GB VRAM on consumer GPUs, under 2 seconds per object, standard .ply output.
Apache 2.0 licensed, with a ComfyUI port already available. It pairs with their Iris-3B model for a prompt → image → 3D pipeline. Setup requires accepting gated access to DINOv3 and FLUX.2-dev VAE encoders on Hugging Face; the 3GB weights auto-download on first run.
More from Multimodal
- FrameForge Reuses Video Model Priors for Unified, Adaptive Conditional Image Generation — FrameXAI · 2026-11-02
- Tencent's Wizard World Model Enables Real-Time Agentic Control of Any Game World — tencent · 2026-11-02
- Open-source ComfyUI node pack renders 2–3 minute 1080P videos with MiniMax H3 in one run — ReferenceFluffy6159 · 2026-11-02
- MiniMax H3 video gen on 8GB RX 6600: 5s clip in 10 minutes with ComfyUI tweaks — posojekh · 2026-11-02
- Generative AI Course Lecture 8: How ComfyUI Custom Nodes Work, From Source to App Mode — CeFurkan · 2026-11-02
- The 'Rian Johnson app' generates strikingly cinematic AI photos — threepointone · 2026-11-02