SuperID by PiktID

SuperID 3.0.1

SuperID is GenAI upscaler, particularly suitable for portrait photos. It is based on GANs, transformers and diffusion models, rendering it extremely powerful in adding missing details.

About

SuperID utilizes generative models to intelligently upscale your photos. It can be extremely powerful in the following scenarios:

• Close-up portrait photos: Since the generative models behind SuperID are trained on human faces, the AI performs better when a single face is provided as input.
• Small thumbnails: When the photo to upscale is at very low resolution (e.g. 128x128), SuperID can add missing details.
• Photos with noise and JPEG artifacts: SuperID can effectively denoise and remove artifacts from your photos.

Getting Started

The following instructions suppose you have already installed a recent version of Python. For a general overview, please visit the API documentation. To use any PiktID API, an access token is required.

Step 0 - Register here. 10 credits are given for free to all new users.

Step 1 - Clone the SuperID library

# Installation commands
$ git clone https://github.com/piktid/superid.git
$ cd superid
$ pip install -r requirements.txt

Step 2 - Export the email and password as environmental variables

$ export SUPERID_EMAIL={Your email here}
$ export SUPERID_PASSWORD={Your password here}

Step 3 - Change in main.py the URL of the image to be upscaled

...
url = 'your-url'
...

Step 4 - Run the main function

$ python3 main.py

Without any additional argument, SuperID utilizes the "super" method and provides the result asynchronously. If you want to upscale x2, use the following general command:

$ python3 main.py --upscaler_type '4' --scale_factor '2' --output_format 'PNG' --flag_email True --prompt '' --prompt_strength '0.35' --controlnet_scale '0.5' --num_inference_steps 20 --seed 0

• upscaler_type: Select which upscaler to use: 0 (None), 1 (Soft Portrait), 2 (Hard Portrait), 3 (Mix), 4 (Best overall)
• upscaling_mode: Select the upscaling mode: fast (takes few seconds, lower quality, not available at the moment), super (takes up to minutes, higher quality overall)
• scale_factor: Select the upscaling factor: upscale time 2 or times 4
• output_format: Save the upscaled image in PNG or JPEG
• flag_email: Get the output link via email once ready
• prompt: Describe your image
• prompt_strength: The lower the more similar to the input image, the higher, the more diverse (range 0-1)
• guidance_scale: How much to follow the description
• controlnet_scale: The higher, the more the upscaling will follow the lines of the input image (range 0-1)
• num_inference_steps: The higher, the more denoising steps
• seed: Choose a seed to replicate the results

Contact

[email protected]