Show HN: A decent text-to-image model trained for ~$20K (Apache 2.0 Weights)
Category: ai-ml
Tags: text-to-image, diffusion-models, pixel-space, apache-2.0, generative-ai
Score: 7.7/10 (Innovation: 8, Technical: 9, Documentation: 7, Utility: 6)
Linum's P-JiT is a 2B-parameter pixel-space text-to-image diffusion model trained for just $21K, using a novel Pyramid-JiT architecture without a VAE. It achieves 4x more aggressive token reduction and converges in 11x fewer samples than their prior model, with Apache 2.0 weights released as a proof-of-concept toward video generation.
Target audience: ML researchers, generative AI engineers, diffusion model practitioners
Repository: https://www.linum.ai/field-notes/pjit-21k
View on Hacker News