Otaku TTI 1.1: sub-1GB anime diffusion model runs in 1.4GB VRAM at 512x512
cynical_hci · reddit · 2026-11-20
Open-sourced Otaku TTI 1.1, a lightweight native 512x512 anime text-to-image latent diffusion model:
- Ultra-light: FP16 weights just 332MB, 1.4GB peak VRAM, runs on 2–6GB laptops and even CPUs
- Fast: 1.14–1.45s per image on an RTX 4050 laptop GPU with DPM-Solver++ (25 steps)
- 4x spatial resolution over v1.0, with richer detail, anime eyes, and layered clothing
- Built-in offline Japanese tag translation (e.g. 初音ミク, 笑顔)
- 166M-param UNet, CLIP ViT-Base/32 encoder, SD-VAE-ft-mse
Trained end-to-end under a 60W power budget on a single laptop, showing high-res generative AI doesn't require datacenter clusters.
More from Multimodal
- Cinematic UFO folklore music video built with Suno and SLOPCANNON — DarkroomSignal · 2026-11-22
- Photographer crafts AI-generated digital muse Nous Girl on faux magazine cover — alexcovo_eth · 2026-11-22
- Turn any Google Maps location into a cinematic drone shot with AI — aziz4ai · 2026-11-22
- Open-source world model PAN 2 wows with GTA-like physics in driving demos — LinusEkenstam · 2026-11-22
- pi-H3 open-sourced: 768p video in under 40s on one RTX 3090 — linoy_tsaban · 2026-11-22
- 2D-to-3D character experiment: hand-drawn illustration becomes consistent 3D render — azed_ai · 2026-11-22