Modifiy images using Instagram-like filters in python. Read how it works on Medium!

pip install git+https://github.com/miroai/instafilter.git@dev

This branch as a few fixes from the original, namely:

• able to run without CUDA
• more flexible training script
• update streamlit app where transformed image can be downloaded at full quality

Example:

from instafilter import Instafilter

model = Instafilter("Lo-fi")
new_image = model("myimage.jpg")

# To save the image, use cv2
import cv2
cv2.imwrite("modified_image.jpg", new_image)

Sample images

Browse samples for every filter in development/examples.

Requirements

Main requirements are opencv, torch==1.7.1 and streamlit (optional for demo app only) and can all be installed via pip install -r development/requirements.txt

Train

To train a filter explicitly:

cd development/train_new_model/
python train.py --input path/to/original.jpg --target path/to/styled.jpg --model_name output_model_name

or passively, the following command will look for transformed images in development/train_new_model/input/ that do not yet have a similarly named model in instafilter/models/ and train new models for those transformed images found:

python train.py

On a g4dn.xlarge one filter takes between 90 minutes to 2 hours to train.

See the code in development/train_new_model or run python development/train_new_model/train.py -h for the full list of options.

Credits

• Made with ❤️ by Travis Hoppe.

• Header image sourced from Will Terra and modified with instafilter.