The data to be translated: PixArt-α is a Transformer-based text-to-image generation model, whose competitive image generation quality and significantly reduced training costs allow it to rival Midjourney and SDXL. With a training strategy decomposition, an efficient T2I Transformer, and high-information-density data training, PixArt-α excels in high-resolution image synthesis and complex text prompts, achieving a training speed that is only 10.8% of Stable Diffusion v1.5. PixArt supports high-resolution image synthesis up to 1024 pixels, reduces training costs by 90%, and offers the AIGC community and startups a new perspective on low-cost, high-quality generative models.