Generative AI for Image and Video Production
summary
Demos tie concrete claims to tools and limits: lyrics are generated with JGBT, music with Zuno AI, images with Midjourney and animated via Runway ML, while Sora and latent diffusion approaches power text‑to‑video experiments. Iterative denoising over 10+ steps and latent patching produce coherent frames but require large memory so prepared videos are shown; trained models predict noise to remove rather than pixels. Ethical and copyright gaps, lawsuits and dataset sourcing recur throughout.
AI-written summary · about
chapters (29)
- 0:00Generative AI for media
- 1:13AI song/video showcase
- 2:23Presenters overview
- 3:07Why generative AI matters
- 4:11Gartner hype cycle reality
- 5:17Text-to-text foundations
- 6:00How GPT predicts text
- 7:04Tokenization and word representation
- 8:32Semantic embeddings
- 9:35Transformer architecture
- 11:47Self-attention explained
- 14:16Text generation steps
- 16:06Text-to-image overview
- 17:30CLIP alignment
- 18:54Diffusion models
- 20:17Why diffusion models
- 23:50Live text-to-image demo
- 26:00Latent diffusion efficiency
- 28:05Text-to-audio techniques
- 29:28Diffusion for audio
- 29:48Video challenges/solutions
- 31:14Attention for video
- 33:03Putting it all together for video
- 34:27Live text-to-video demo
- 35:29Advanced video apps
- 36:57Video AI limitations
- 38:25Ethics in generative AI
- 40:11Environmental/labor impacts
- 41:14Frequently asked questions
related talks
17mThe State of Generative Media
53mGenerative AI: Overview and use cases
14mGenerative AI models: Practical Business Applications
01h 27mImage Generation with AI
01h 14mLatest trends in AI
32mAI Video: The Future of Imagination and Production
43mEncounters of the Third Kind With the Multimodal Generative AI Revolution
40mThe Marvelous Magic of Multimodal AI