The Old Models Foundation has launched a website and associated organization that gives the public interactive access to earlier generations of image-generation artificial intelligence. The project aims to preserve and exhibit the models and visual effects that characterized the early phases of image-based AI.
Which models are available
The collection spans several decades of work, from the Efros–Leung texture synthesis (1999) and image quilting (2001) through variational autoencoders (VAE) in 2013, to notable models from 2015–2018 and more recent systems from 2021–2023. Examples listed by the Foundation include:
- Efros–Leung texture synthesis (1999) and image quilting (2001)
- VAE (2013)
- DRAW, DeepDream, AlignDRAW, DCGAN (2015)
- pix2pix (2016)
- AttnGAN (2017)
- BigGAN, Ganbreeder, StyleGAN (2018)
- VQGAN + CLIP, DALL·E Mini, Disco Diffusion (2021)
- LAION-400M, DALL·E Mega, Stable Diffusion (2022)
- ControlNet, SDXL 1.0 (2023)
The list illustrates how techniques and output quality evolved and highlights the models that were significant for hobbyist and experimental users.
Usage rules and limits
Visitors to the site can experiment with these models, but the platform enforces a usage limit: accounts are allowed up to 25 generations per day. This cap is intended to control load and manage platform usage.
Why these old models matter
The visual defects and peculiar aesthetics of early image generators are historically informative: many early systems produced distinctive failures and styles that became part of their signature look, and these characteristics are often difficult for modern, better-performing models to replicate. As musician Brian Eno put it, “Whatever you now find weird, ugly, uncomfortable and nasty about a new medium will surely become its signature.” That observation helps explain why revisiting early models is valuable — their quirks and limitations can have aesthetic importance and illustrate the field’s development.
Where to try them
The models and interactive tools are available on the Old Models Foundation’s official website, where visitors can run and compare different systems and see firsthand how generative image AI has changed over time.
The collection provides a hands-on way to understand how hobbyists and researchers experienced image generation during its formative years and preserves visual artifacts from AI’s recent history.
