Otaku TTI 1.0: a 318MB anime diffusion model that runs on under 1.2GB VRAM
cynical_hci · reddit · 2026-11-18
A developer released Otaku TTI 1.0, a latent diffusion model for anime character art trained from scratch on a laptop RTX 4050. Key specs: 166M-parameter latent UNet with CLIP cross-attention, 318MB FP16 download, under 1.2GB VRAM at inference (CPU-only works in 3-5s), 190-220ms per image on entry GPUs with DPM-Solver++, fine-tuned 173k+ steps on curated Danbooru art. Licensed CC-BY-NC 4.0, available on GitHub and Hugging Face with a one-command quickstart.
More from Multimodal
- AI video of the day: a cat calmly eating ramen amid total chaos, made with seedance 2.5 — aftahi_ai · 2026-11-18
- Same prompt, same Mac: MiniMax H3 generates in 140s vs Qwen's 15 minutes — FinanceYF5 · 2026-11-18
- 4-minute "slopcore" music video with song and footage generated in one pass — FinanceYF5 · 2026-11-18
- Whole room Gaussian-splatted then relit in Blender — watch the shadows move — FinanceYF5 · 2026-11-18
- AI popstar Claudia stars in 'Bring Jensen Back' parody video made with Dreamina, Hailuo, Suno — AIandDesign · 2026-11-18
- The viral influencer doesn't exist: Arcads AI workflow ships UGC ads in minutes — aftahi_ai · 2026-11-18