uniworld-semantic-vision

Combine semantic encoders from multimodal LLMs with contrastive learning to create unified high-resolution encoders for both visual understanding and generation tasks without relying on VAE compression.

Independent GitHub snapshot · Automated classification · Original documentation governs