Alibaba's Z-Image paper: a 6B image model trained for under $630K, runs on 16G VRAM
teortaxesTex · x · 2026-10-28
teortaxesTex praises Alibaba's Z-Image team as "whale-tier outliers" behind one of the most beautiful things to come out of China. The paper's conclusion reveals: Z-Image is a 6B-parameter S3-DiT model trained in just 314K H800 GPU hours at under $630K total cost, delivering top-tier photorealism and bilingual text rendering; Z-Image-Turbo does sub-second inference and fits in 16G consumer VRAM, plus an efficiently derived Z-Image-Edit model.
Related event: Alibaba's Z-Image Team Reportedly Disbanded(2 posts)→
More from Multimodal
- Genie 4 steps into old photos: researcher revisits a lost memory of his late mom — holynski_ · 2026-10-28
- Abstract Visuals on an RTX 4070: The Case for LTX Over MiniMax — ART-ficial-Ignorance · 2026-10-28
- Zero-AI-at-runtime design generator hits run 8, new aesthetics expected by run 10 — Afinetheorem · 2026-10-28
- Painting with EmbeddingGemma 2: ~30 images/sec on Apple Neural Engine after 12h A100 training — GlennCameronjr · 2026-10-28
- Krea2 BBOX models released: 40k-image fullrank finetune with multi-character LoRA support — Amazing_Painter_7692 · 2026-10-28
- One-line Tekken-style prompt in HeyGen Video yields ad-grade 2K fight footage — aziz4ai · 2026-10-28