Linum V3 - GoKawiil

Linum's new JiT-DDT model cuts text-to-image training time 3.6x

Linum has released JiT-DDT, an experimental encoder-decoder architecture that trains text-to-image generation models using 3.6 times fewer GPU-hours than its previous Linum v2 system, while producing images with four times the pixel resolution. The approach merges compression and generation into a single pixel-space model rather than relying on separate VAE and diffusion transformer components, addressing a detail-loss problem seen in earlier pixel-space designs. Code and weights are being released under Apache 2.0 as a research checkpoint on the way to Linum v3.

linum.ai · 2026-09-16

Linum previews v3 roadmap, betting on data quality over raw scale for video AI

Linum has begun detailing its work on Linum v3, the successor to its open-weight text-to-video model released in January, in a new blog series. The company says its focus is on three goals: improving how well outputs follow prompts, speeding up training and inference, and making generated physics more consistent, while arguing that most recent progress in image and video generation comes from better data practices rather than new model architectures.

linum.ai · 2026-08-27