min-dalle-test/README.md

65 lines
2.3 KiB
Markdown
Raw Normal View History

2022-06-27 19:46:04 +00:00
# min(DALL·E)
2022-06-27 17:38:35 +00:00
2022-07-01 23:12:43 +00:00
[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/kuprel/min-dalle/blob/main/min_dalle.ipynb)
 
2022-06-29 19:24:09 +00:00
[![Replicate](https://replicate.com/kuprel/min-dalle/badge)](https://replicate.com/kuprel/min-dalle)
2022-07-01 23:12:43 +00:00
 
[![Join us on Discord](https://img.shields.io/discord/823813159592001537?color=5865F2&logo=discord&logoColor=white)](https://discord.gg/xBPBXfcFHd)
2022-06-27 21:26:05 +00:00
2022-07-02 12:38:45 +00:00
This is a minimal implementation of Boris Dayma's [DALL·E Mini](https://github.com/borisdayma/dalle-mini) in PyTorch. It has been stripped to the bare essentials necessary for doing inference. The only third party dependencies are numpy, requests, pillow and torch.
2022-07-02 16:32:54 +00:00
To generate a 3x3 grid of DALL·E Mega images it takes
- **35 seconds** with the standard GPU runtime in Colab
- **15 seconds** with an A100 on replicate
2022-06-27 17:51:48 +00:00
2022-07-01 23:19:02 +00:00
The flax model and code for converting it to torch can be found [here](https://github.com/kuprel/min-dalle-flax).
2022-07-01 23:03:18 +00:00
## Install
2022-06-27 17:51:48 +00:00
2022-07-01 22:52:40 +00:00
```bash
2022-07-01 22:30:22 +00:00
$ pip install min-dalle
```
2022-06-27 21:26:05 +00:00
2022-07-01 23:03:18 +00:00
## Usage
2022-06-27 21:28:40 +00:00
2022-07-01 23:03:18 +00:00
### Python
2022-07-01 23:02:49 +00:00
2022-07-02 00:27:29 +00:00
Load the model parameters once and reuse the model to generate multiple images.
2022-07-01 22:30:22 +00:00
2022-07-01 22:32:48 +00:00
```python
from min_dalle import MinDalle
2022-07-01 22:30:22 +00:00
2022-07-02 00:24:19 +00:00
model = MinDalle(is_mega=True, models_root='./pretrained')
2022-06-27 19:46:04 +00:00
```
2022-07-02 00:25:49 +00:00
The required models will be downloaded to `models_root` if they are not already there. Once everything has finished initializing, call `generate_image` with some text and a seed as many times as you want.
2022-07-01 22:54:05 +00:00
```python
2022-07-02 13:26:30 +00:00
text = 'a comfy chair that looks like an avocado'
2022-07-01 23:08:42 +00:00
image = model.generate_image(text)
display(image)
2022-06-27 17:38:35 +00:00
```
2022-07-02 03:39:48 +00:00
![Avocado Armchair](https://github.com/kuprel/min-dalle/raw/main/examples/avocado_armchair.png)
2022-06-27 17:38:35 +00:00
```python
2022-07-02 13:26:30 +00:00
text = 'court sketch of godzilla on trial'
image = model.generate_image(text, seed=6, grid_size=3)
display(image)
2022-06-27 17:38:35 +00:00
```
2022-07-02 13:26:30 +00:00
![Godzilla Trial](https://github.com/kuprel/min-dalle/raw/main/examples/godzilla_trial.png)
2022-07-01 23:15:31 +00:00
### Command Line
2022-07-01 23:19:02 +00:00
Use `image_from_text.py` to generate images from the command line.
2022-07-01 23:15:31 +00:00
```bash
$ python image_from_text.py --text='artificial intelligence' --seed=7
```
2022-07-02 03:39:48 +00:00
![Artificial Intelligence](https://github.com/kuprel/min-dalle/raw/main/examples/artificial_intelligence.png)
2022-07-01 23:15:31 +00:00
```bash
2022-07-02 13:26:30 +00:00
$ python image_from_text.py --text='trail cam footage of gollum eating watermelon' --mega --seed=1 --grid-size=3
2022-07-01 23:15:31 +00:00
```
2022-07-02 13:32:15 +00:00
![Gollum Trailcam](https://github.com/kuprel/min-dalle/raw/main/examples/gollum_trailcam.png)