LLMs as directors: a curated reading list for multimodal generation

What it does This repository is a curated bibliography—essentially a well-organized reading list—charting the intersection of large language models and multimodal generation. It catalogs papers across image, video, 3D, and audio generation and editing, splitting each section into LLM-based and non-LLM-based approaches. The list also branches into related territory like multimodal agents, understanding, and safety.

The interesting bit Rather than dumping links, the authors impose a strict taxonomy: generation versus editing, LLM-driven methods versus legacy CLIP/T5 pipelines, plus cross-cutting tags like tokenizer and customization. It serves as the companion repo to an arXiv survey paper, so the curation carries the weight of a formal literature review.

Key highlights

• Covers four generation modalities—image, video, 3D, and audio—plus editing tasks for each.
• Splits literature into LLM-driven methods and older non-LLM backbones for quick comparison.
• Includes adjacent topics often left out of pure generation surveys: multimodal agents, understanding, and safety.
• Searchable by author name or tags such as customization, interactive, and tokenizer.
• Explicitly welcomes pull requests to add new papers.

Caveats

• This is a bibliography, not a framework: expect paper titles and arXiv badges, not runnable code or reproduced results.

Verdict Researchers and engineers drowning in multimodal-generation literature should bookmark this; if you are looking for model weights or turnkey training scripts, look elsewhere.