Hugging Face trains competitive text-to-image model in 24 hours for $1500
hugging face trained a competitive text-to-image diffusion model in just 24 hours using a $1500 compute budget. the 'prx part 3' speedrun utilized x-prediction to train directly in pixel space without a vae, starting directly at 512px resolution. they also incorporated l1pips and dino-based perceptual losses to boost convergence speed and visual quality. the team open-sourced the training code and experimental framework on github.
Hugging Face demonstrated a 24-hour text-to-image training speedrun with a total compute budget of $1,500. The team combined architectural optimizations, including the x-prediction formulation to train directly in pixel space, eliminating the need for a VAE. This approach allowed them to start training directly at 512px and fine-tune at 1024px using a patch size of 32. They integrated auxiliary perceptual losses—specifically LPIPS and a DINOv2-based loss—applied at all noise levels to improve final image quality. The team released the training code and framework on GitHub under an open-source license.