PRX Part 3 — Training a Text-to-Image Model in 24h!
The PRX team released a 24‑hour training run that combines pixel‑space diffusion, token routing with TREAD, perceptual losses (LPIPS and DINO), representation alignment via REPA, and the Muon optimizer. Starting at 512px and fine‑tuning at 1024px, the model achieves usable image quality in a single day on modest hardware. The code and configs are open‑source on GitHub, enabling reproducibility and further experimentation. The experiment demonstrates how modern engineering can reduce diffusion training costs from millions to a day’s worth of compute.
The Core Development
In the last two posts ( Part 1 and Part 2 ), we explored a wide range of architectural and training tricks for diffusion models. We tried to evaluate each idea in isolation, measuring throughput, convergence speed, and final image quality, and tried to understand what actually moves the needle.
In this post, we want to answer a much more practical question:
Instead of optimizing one dimension at a time, we’ll stack the most promising ingredients together and see how far we can push performance under a strict compute budget.
How It Works & Technical Architecture
These losses are lightweight compared to the main transformer forward pass, and in our setup they add only a small overhead while providing a consistent quality boost.
To make each step cheaper, we use token routing with TREAD [ Krause et al., 2025 ]), which randomly selects a fraction of tokens and lets them bypass a contiguous chunk of transformer blocks, then re-injects them later so nothing is dropped.
Following the TREAD recipe, we route 50% of the tokens from the 2nd block to the penultimate block of the transformer.
Concretely, we apply the alignment loss once,at the 8th transformer block with a loss weight of 0.5 .
Strategic Perspective & Ecosystem Impact
This is very far from the early diffusion days, where training competitive models could cost millions of dollars. The goal here is to demonstrate how much the field has evolved and how far careful engineering can take you in just a single day of training.
This speedrun is not just a fun experiment. It will likely serve as the foundation for our large-scale training recipe going forward.
Alongside the results, we’re also open-sourcing our code ( Github link ), which contains:
Why it matters
Showing that a competitive text‑to‑image model can be trained in 24 hours on modest hardware lowers the barrier to entry for research labs and accelerates innovation in diffusion technology.