afiaka87 / pyglide

The predecessor to DALLE-2, GLIDE (filtered) with faster PRK/PLMS sampling.

  • Public
  • 18.6K runs
  • GitHub
  • Paper
  • License
Run with an API
Playground API Examples README Versions

Examples

View more

Run time and cost

This model costs approximately $0.042 to run on Replicate, or 23 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia T4 GPU hardware. Predictions typically complete within 4 minutes. The predict time for this model varies significantly based on the inputs.

Readme

pyglide

OpenAI filtered all humans out of the training set for this model. This means a lot of prompts simply can’t work. Still a fun model though and incredibly fast!

Thanks to “Pseudo Numerical Methods for Diffusion Models on Manifolds”, results are now even faster and more accurate. Credit goes to Katherine Crowson for implementing this for GLIDE specifically (https://github.com/crowsonkb/glide-text2im).

an analog clock hanging on a blue wall an analog clock hanging on a blue wall

a lonely robot in the middle of the field a lonely robot on hanging out on a cliff

a goose made of paper. paper goose. a goose rendered in minecraft. minecraft goose.

Installation

First clone this repository:

git clone https://github.com/afiaka87/text-glided-diffusion.git
cd text-glided-diffusion

You also need to install glide-text2im from openai’s repository.

python3 -m venv .venv
source .venv/bin/activate
(.venv) python -m pip install -r requirements.txt
(.venv) git clone https://github.com/openai/glide-text2im.git
(.venv) cd glide-text2im/
(.venv) python -m pip install -e .
(.venv) cd ../

Usage

time python tgd.py --prompt "the beach at sunset"
Selected device: cuda:0.
1. Creating model and diffusion.
1. Done.
2. Running base GLIDE text2im model.
2. Base model generations complete. Check glide_outputs/base/the_beach_at_sunset/the_beach_at_sunset.png for generations.
3. Loading GLIDE upsampling diffusion model.
3. Done.
4. Running GLIDE upsampling from 64x64 to 256x256.
4. Done. Check glide_outputs/sr/the_beach_at_sunset/the_beach_at_sunset.png for generations.

real    1m4.775s
user    1m9.648s
sys     0m8.894s

Detailed Usage

usage: tgd.py [-h] --prompt PROMPT [--batch_size BATCH_SIZE] [--guidance_scale GUIDANCE_SCALE] [--base_x BASE_X] [--base_y BASE_Y] [--respace RESPACE] [--prefix PREFIX] [--upsample_temp UPSAMPLE_TEMP]

optional arguments:
  -h, --help            show this help message and exit
  --prompt PROMPT       a caption to visualize
  --batch_size BATCH_SIZE
  --guidance_scale GUIDANCE_SCALE
  --base_x BASE_X       width of base gen. has to be multiple of 16
  --base_y BASE_Y       width of base gen. has to be multiple of 16
  --respace RESPACE     Number of timesteps to use for generation. Lower is faster but less accurate.
  --prefix PREFIX       Output dir for generations. Will be created if it doesn't exist with subfolders for base and upsampled.
  --upsample_temp       0.0 to 1.0. 1.0 can introduce artifacts, lower can introduce blurriness.