A new text-to-image model called P-JiT 21K has been released, featuring 2 billion parameters and trained for approximately $21,000. This model operates directly in pixel space, eliminating the need for a VAE. It has been ported to ComfyUI as custom nodes, offering optimized versions like bf16 and int8 for reduced file size and faster performance, and can run with as little as 4GB of VRAM. AI
IMPACT Provides a new option for pixel-space text-to-image generation with optimized performance and low VRAM requirements.
RANK_REASON Release of a new AI model with technical details and community porting. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →