Lucidly | Widening human visual perspective with AI

Project Statement
Great stories often rely on story tellers. Often, these tellers converge the meaning to a singular expression and hence constrain perspectives and ultimately creativity. This project aims at widening the perspective lens through AI (stable-diffusion model) by creating multiple interpretations / visualizations off a text input. All these would be smoothly compiled as a GIF.

I’ve been working a bit on this, here are some premature results

The How

  • Prompt engineer story/text for the model.
  • Create an interpolation module to traverse and generate across multiple perspectives in the model’s latent space.
  • Create a function to smoothen transitions between two consecutive frames.
  • Compile it into a GIF / video.

Creating an open-source colab notebook walkthrough for generating perspectives with stable diffusion.

Amit Singh (metamyth)
R&D, Paperplane Technology
Researcher, Active Inference Lab

Discord: metamyth #8558
USDC Wallet Address: 0x0159af752e0220ed3eef439bef36f982cc0a6fbf

This is an interesting idea and I’m not sure if you are referring to a perspective in the sense of a different semantic interpretation (different understanding/meaning) or to a different artistic style ?

Thanks. I meant the former. As the model understands words differently everytime, we’ll have different interpretations in it’s latent space. That’s what I’m looking to explore here

It might be interesting to seed the next frame with part of the current frame.