A Survey of Generative Categories and Techniques in Multimodal Generative Models

Name: A Survey of Generative Categories and Techniques in Multimodal Generative Models
Uploaded: 2026-10-22T15:18:10.071300+00:00
Duration: 1388 s
Description: This survey maps Multimodal Generative Models (MGMs) across six types, including text-to-video and 3D generation. It details foundational techniques like Self-Supervised Learning and Chain-of-Thought prompting used to create diverse outputs. The paper proposes a unified evaluation framework focusing on faithfulness, compositionality, and robustness to guide future research.

This survey maps Multimodal Generative Models (MGMs) across six types, including text-to-video and 3D generation. It details foundational techniques like Self-Supervised Learning and Chain-of-Thought prompting used to create diverse outputs. The paper proposes a unified evaluation framework focusing on faithfulness, compositionality, and robustness to guide future research.

Open the full episode