The Idol image to 3D model now finishes in about 0.75 seconds on an H200
This is a design post classified by AGK Intelligence as AI visuals for design (a launch), kept by the Design Radar because it carries real work, not commentary.
Idol Turbo The Idol image to 3D model now finishes in about 0.75 seconds on an H200 This is a faster version of Idol, the single image to 3D Gaussian model I shared earlier. It still takes one object photo and outputs up to a million 3D Gaussians. Idol needs 50 steps with guidance, which means 100 network passes. Idol Turbo does it in 5 steps without guidance, so 20 times fewer passes. Two small LoRA experts split the work between the noisy early steps and the cleaner late steps. The base model weights stay untouched, and the training set was 93 single object images. It works in ComfyUI
Posted by DEXTUNE (1.4k followers) 4 days ago · 60 likes · 2.3k views · view the original post on X. Kept by the Design Radar as AI visuals for design. Tools mentioned: idol-turbo.
More design work like this
- THE WOLVES HAVE A NEW CREATIVE TOOL 🐺 — @CryptoEmpressX
- YOU DON'T NEED A BLANK CANVAS TO MAKE A GOOD AD — @DataChaz
- GPT Image 2.6 by @OpenAI is now on Sunra. — @sunra_ai
- Another hybrid style. This time I used it for science. — @Gdgtify
- From a static movie poster to a cinematic reveal 🐉🔥 — @LudovicCreator
- Some animations are simply filled with beauty. — @Artedeingenio
- Qwen Qoduck（Qwen 4?） 超惊艳 3D 白宫建模测试作品欣赏： — @leo114119
- 💥细节感拉满！Qwen 4.0 蜜蜂向日葵测试炸了！ — @NFT_Chen
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 13.9k posts from 5k X accounts over the last 21 days, 2.6k tools, 17 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-08 18:48 UTC. Full method.