File size: 3,783 Bytes
d4779c4 3b7350b d4779c4 3b7350b d4779c4 38dc412 d4779c4 38dc412 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 |
# min(DALL路E)
[](https://colab.research.google.com/github/kuprel/min-dalle/blob/main/min_dalle.ipynb)
This is a fast, minimal port of Boris Dayma's [DALL路E Mega](https://github.com/borisdayma/dalle-mini). It has been stripped down for inference and converted to PyTorch. The only third party dependencies are numpy, requests, pillow and torch.
To generate a 4x4 grid of DALL路E Mega images it takes:
- 89 sec with a T4 in Colab
- 48 sec with a P100 in Colab
- 13 sec with an A100 on Replicate
The flax model and code for converting it to torch can be found [here](https://github.com/kuprel/min-dalle-flax).
## Install
```bash
$ pip install min-dalle
```
## Usage
Load the model parameters once and reuse the model to generate multiple images.
```python
from min_dalle import MinDalle
model = MinDalle(
models_root='./pretrained',
dtype=torch.float32,
is_mega=True,
is_reusable=True
)
```
The required models will be downloaded to `models_root` if they are not already there. Set the `dtype` to `torch.float16` to save GPU memory. If you have an Ampere architecture GPU you can use `torch.bfloat16`. Once everything has finished initializing, call `generate_image` with some text as many times as you want. Use a positive `seed` for reproducible results. Higher values for `log2_supercondition_factor` result in better agreement with the text but a narrower variety of generated images. Every image token is sampled from the top-$k$ most probable tokens.
```python
image = model.generate_image(
text='Nuclear explosion broccoli',
seed=-1,
grid_size=4,
log2_k=6,
log2_supercondition_factor=5,
is_verbose=False
)
display(image)
```
<img src="https://github.com/kuprel/min-dalle/raw/main/examples/nuclear_broccoli.jpg" alt="min-dalle" width="400"/>
credit: [https://twitter.com/hardmaru/status/1544354119527596034](https://twitter.com/hardmaru/status/1544354119527596034)
### Saving Individual Images
The images can also be generated as a `FloatTensor` in case you want to process them manually.
```python
images = model.generate_images(
text='Nuclear explosion broccoli',
seed=-1,
image_count=7,
log2_k=6,
log2_supercondition_factor=5,
is_verbose=False
)
```
To get an image into PIL format you will have to first move the images to the CPU and convert the tensor to a numpy array.
```python
images = images.to('cpu').numpy()
```
Then image $i$ can be coverted to a PIL.Image and saved
```python
image = Image.fromarray(images[i])
image.save('image_{}.png'.format(i))
```
### Interactive
If the model is being used interactively (e.g. in a notebook) `generate_image_stream` can be used to generate a stream of images as the model is decoding. The detokenizer adds a slight delay for each image. Setting `log2_mid_count` to 3 results in a total of `2 ** 3 = 8` generated images. The only valid values for `log2_mid_count` are 0, 1, 2, 3, and 4. This is implemented in the colab.
```python
image_stream = model.generate_image_stream(
text='Dali painting of WALL路E',
seed=-1,
grid_size=3,
log2_mid_count=3,
log2_k=6,
log2_supercondition_factor=3,
is_verbose=False
)
for image in image_stream:
display(image)
```
<img src="https://github.com/kuprel/min-dalle/raw/main/examples/dali_walle_animated.gif" alt="min-dalle" width="300"/>
### Command Line
Use `image_from_text.py` to generate images from the command line.
```bash
$ python image_from_text.py --text='artificial intelligence' --no-mega
```
<img src="https://github.com/kuprel/min-dalle/raw/main/examples/artificial_intelligence.jpg" alt="min-dalle" width="200"/>
<br />
[Sponsor this work](https://github.com/sponsors/kuprel) |