There are two implemented servers: frontend and backend. Frontend service uses streamlit to develop an web app. Backend uses flask and diffusers to generate images. For personalized Kim generation I used text inversion technique. Token was taken from civitai.com. Several photorealistic checkpoints of stable diffusion were used but best results are get with this one. Here is an example of a generation
You can check other examples under imgs/ folder
you can either build image from scratch using:
docker compose build && docker compose up -d
or use prebuilt images by simply running
docker compose up -d
In both cases web app will be available under localhost:8501
Approximate VRAM usage is 6GB. Docker image is build only supporting gpu usage.