Project Statement
[Results from last model] (Lucidly – Google Drive ).
Stable Diffusion models serve as a rich medium for creative expression. Each prompt corresponds to a token-set in the text-encoder’s latent space and then produces static images for a singular prompt well. When storytelling or creating visual motion from A → B based from prompts, both scenarios create independent generations and hence lack cohesion between them. Text models solved incoherence by implementing a function in the Decoder sampling strategy - balancing novelty in generations and conditional dependence on previously generated sequences using hyperparamter knobs.
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