splat-repair LoRA updated: fixing Gaussian splat artifacts with Qwen Image Edit
Spy4killer · reddit · 2026-10-28
The author updated their splat-repair LoRAs for Qwen-Image-Edit 2511 (three trained LoRAs total) and rebuilt the old GaussianViewer ComfyUI node to be faster and more efficient. The 'fix it' LoRA repairs Gaussian splat artifacts with just an activation word, though better prompting helps; it also works on people. Everything is open-sourced on GitHub, crediting the original idea by wuli-DX and CarlMarkswx.
More from Multimodal
- Google DeepMind scientist's reply reveals AI VFX credits on F1, Alien: Romulus and Furiosa — TomLikesRobots · 2026-10-28
- Google researcher studies whether image models like Nano Banana can look things up — YonatanBitton · 2026-10-28
- Genie 4 demo: an AI-generated jigsaw puzzle — CodeByPoonam · 2026-10-28
- Genie 4 demo: stylized 3D horse animation — CodeByPoonam · 2026-10-28
- Genie 4 hands-on: playable game generated in minutes impresses users — CodeByPoonam · 2026-10-28
- AI-generated indie pop song 'Loopin'' shows off full arrangement structure — Cyborgized · 2026-10-28