diffusion.pdf; a PDF that draws Pictures logo

diffusion.pdf; a PDF that draws Pictures

Show HN: Diffusion PDF – A Diffusion Image Model Embedded Entirely in a PDF File

Visit Website

At a glance

Explore the listed capabilities, then check the details that matter to you.

  • AI assistant
  • Natural language interface
  • Workflow automation
show-hnLLMaiai assistantautomation
diffusion.pdf; a PDF that draws Pictures screenshot 1

Overview

diffusion.pdf is a PDF file that generates images using an embedded diffusion image model. It is a proof-of-concept that runs a denoising diffusion model inside a PDF, similar to Stable Diffusion, Midjourney, or Dall-e but described as way worse. The model was trained on about 130 categories from Google's Quick Draw dataset of doodle images and recognizes 131 known objects; inputs outside those objects are mapped to a known object poorly. The PDF requires Chrome or another Chromium browser, with no Safari or Firefox support, and at least 75MB of storage. It renders generated drawings as ASCII art using 28 rows of text because of PDF form fields, and the generation is inverted to show against a white background. The file is 75MB, has 55,403,136 int8 parameters, uses 886M MAC per image, and runs at about 15.3 seconds per image at 58M MAC/s. It is built by Alex Van Doren and is a fork of Evan Zhou's llm.pdf.

Pricing

Detailed plans have not been confirmed in our catalog. Check the official website for current limits and billing terms.

Visit Website

Prices and limits may change. Confirm the currency, billing period, seat minimum and usage caps on the official website.

Use Cases

  • Users use diffusion.pdf to run a diffusion image model fully inside a PDF.
  • Users use diffusion.pdf to generate images from one of 131 known objects.
  • Users use diffusion.pdf to view generated drawings as ASCII art in PDF form fields.
  • Users use diffusion.pdf in Chrome or another Chromium browser.
  • Users use diffusion.pdf to generate an image in about 15.3 seconds at 58M MAC/s.
  • Users use diffusion.pdf by downloading the 75MB PDF.

Features

AI assistant

Natural language interface

Workflow automation

Reviews

No reviews yet. Be the first to share your experience!

FAQ

So is this AI?

The site says it is technically a denoising diffusion model similar to Stable Diffusion, Midjourney, or Dall-e, but much worse. It was trained on about 130 categories from Google's Quick Draw dataset of doodle images.

So how fast is it?

The site says it takes about 1.8 seconds per step when generating in Chrome on an M5 Macbook Air, and the specs list a speed of about 15.3 seconds per image at 58M MAC/s.