Lucidly: Text and Image Art Symposium

Proposals Round 7
#1

Project Statement
One statement: Letting the model generate its own story through prompt generation with context and sound meaning.
My last iteration on this project aimed at generating multiple interpretations off a text prompt and engineering animations between them smoothly compiled as GIFs. This one would focus on the research of how to make the model interrogate itself, i.e generate prompts for itself on the go and in context with what happened before it. Core to the research would be finding a way mathematically to generate cohesive meaningfulness to human mind.

[My results from last model] (Lucidly – Google Drive). They were generated with one liner prompts.

Deliverables
Research report and Analysis on how to generate coherent text prompts and hence meaningful visual stories with animation.

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

Twitter: https://twitter.com/not_amyth?t=BSpbbssQAIkrRAgg1ZTUIQ&s=09
Discord: metamyth #8558
USDC Wallet Address: 0x0159af752e0220ed3eef439bef36f982cc0a6fbf

4 Likes
#2

This sounds really exciting. It can really help in interpreting the model’s predictions. Did you plan any experiments and test benchmark to test you approaches?

Proposal: Human Evaluation of Methods for fine-tuning large language models
#3

Yeah sure it does. My approaches will measure context in consecutive frames. You can think of this in transformer vocab as attention over frames/model outputs. More to be worked for sure, would love to know if you have something in mind.