See https://docs.google.com/document/d/1U82LQtQbydvW8vT86GGD8m5-kCdKCZ2rEBD9VlVugUs/edit?usp=sharing

In short:
Project OpenDiffusion: Reimplementing and improving stream diffusion

While the proposed stream diffusion greatly improves speed over other diffusion approaches which makes interactivity and live generation possible, it also severely reduces creative freedom and is constrained to diffusing incoming video. The initial stream diffusion project is also an introduction to a concept, rather than a mature implementation of the potential of the concept. In this project, we aim to reimplement the stream diffusion concepts from the paper into “OpenDiffusion”.

In this first POC, we will experiment with different changes to the pipelines to enable more creative freedom at the expense of speed and efficiently,  but retain the continuous streaming nature. Additionally, introducing Prompt Arrangement Clips (PAC) as a mechanism to drive changing text prompt input and other controls and variables to steer continuous generation in “dream mode” or otherwise. The implementation re-introduces negative prompts into continuous diffusion.

Budget: $1,000

Posted On: February 17, 2024 06:52 UTC
Category: Full Stack Development
Skills:Python, JavaScript, Stable Diffusion, Machine Learning

Country: Netherlands

click to apply

Powered by WPeMatico