DINOv2
HoldTechniques
A self-supervised vision foundation model used here as a perceptual feature extractor.
Why it's here
Placed in Hold: 1 article(s) of evidence from 1 source(s), led by research-stage coverage, with 0 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 7Hugging Face Blog·3/3/2026researchPhotoroom shares a 24-hour text-to-image training recipe
Photoroom describes how it combined several diffusion-model training optimizations into a 24-hour speedrun on 32 H200 GPUs with a roughly $1,500 compute budget. The recipe uses x-prediction in pixel space, starts at 512px before fine-tuning at 1024px, and adds perceptual losses such as LPIPS and DINOv2-based supervision. The team also open-sourced the code and experimental framework used in the run.