Repo for applications created with streamlit frontend and modal as the backend
StreamlitModal.mp4
The frontend has been created using streamlit that has a simple authenticator. The user can enter a prompt and change the number of generation steps, guidance scale and sequence length. It also allows the possibility of entering a manual seed number for image consistency, LORA path and file name for particualr style. The UI makes an API request to a serverless gpu infrastructure (Modal) that host the open source text to image model - Flux. The Modal server takes in the information, generates an image and sends it back as JSON response and is displayed to user. More details about the project can be found here and here
CogVideoX.1_5B.Text2Video.Demo.mp4
A frontend has been developed with streamlit where the user can enter a prompt describing the scene they want to describe in a sequence. It has the option of entering number of inference steps, guidance scale, manual seed. There is an API created using modal container that takes in these values and generates a video. The video is saved on the modal container and also sent back to the streamlit application as base64 encoded content. An open source model CogVideoX 1.5B is used to generate the video. The application can further be enhanced by adding an authentication component.
- Add Websockets and display video generation progress on the streamlit application
1739496488_the_ultrawideangle_lens_follows_closely_from_the_hood_with_raindrops_continuously_splattering_agains_1.mp4
SampleOutput.mp4
A streamlit frontnend that allows to perform Text to Video generation where the user can enter prompt in a sequential manner describing the sequence. The open source Hunyaun model runs on a modal container with A100-80 HB GPU that takes in prompt and generation parameters set by user. Although Modal inherently doesn't allow the possibility to obtain progress bar and possibility to cancel an exisitng request, these features have been incorporated by designing separate end points and modal containers. The GUI shows the user updates and progress made on the modal container as the generation takes substantial time to complete.
A frontend developed on streamlit with authentication capabilities. Using serverless gpu service Modal applications running DeepSeek dsitlled models ( Llama-8B, Qwen-14B, Qwen-32B, Llama-70B ) with webpoint api. Frontend allows to seamlessly switch between one model to other. There is some model loading time involved on model switching. Model parameters - temperature, top_p, top_k, max_new-token, manual_seed can be varied on each request. The distilled models offer very good reasoning capabilities on each request.