Today in Generative Media

1000 subs! AI with emotions; AI newscaster bots; Smart but sluggish Strawberry?

As of 7:30 this morning this newsletter now has 1000 subscribers! Thanks everyone!

[Image created using FLUX.1 [schnell] on Replicate.]

News

• This New Tech Puts AI In Touch With Its Emotions—and Yours (Wired)

• An AI Bot Named James Has Taken My Old Job (Wired)

• This Week in AI: OpenAI’s new Strawberry model may be smart, yet sluggish (TechCrunch)

• Opinion: ‘He surprised me’: Oprah on her new AI special with Sam Altman (Washington Post)

• OpenAI gives artists access to unreleased tools like Sora for New York gallery exhibit (cnbc.com)

• OpenAI is reportedly raising at a $150 billion valuation — making it worth more than 88% of S&P 500 firms (Business Insider)

• Inside the Heated Controversy That’s Tearing a Writing Community Apart (Slate)

• Opinion Letters: Was It Written by ChatGPT, or by a Novelist? (New York Times)

Events

• AI-nimation: Reshaping the Animation Industry. Thursday, Oct 17. 5:00pm – 9:00pm (LA Tech Week)

• Saturday, September 14: Co-creating with Art and AI end of residency exhibition vernissage 5@7 (Co-Creating with Art and AI). Montreal, Quebec.

Software

• Google’s AI note-taking app NotebookLM can now explain complex topics to you out loud (TechCrunch)

• Mistral releases Pixtral 12B, its first multimodal model (TechCrunch)

• Meet Verse, an AI-powered creative app that helps Gen Z design and publish expressive content (TechCrunch)

• LLaMA-Omni: The open-source AI that’s giving Siri and Alexa a run for their money (VentureBeat)

• Hi3D: Pursuing High-Resolution Image-to-3D Generation with Video Diffusion Models (GitHub) Paper on ArXiv.

• Fluxgym: Train FLUX LoRA LOCALLY with LOW VRAM (X)

• Flux from @bfl_ml is now on RenderNet! Create hyper-realistic, character-consistent images with just one reference image — no LORAs required. (X)

• Chat With Your Docs (GitHub)

• PaperQA2 is a package for doing high-accuracy retrieval augmented generation (RAG) on PDFs or text files, with a focus on the scientific literature. See our recent 2024 paper to see examples of PaperQA2's superhuman performance in scientific tasks like question answering, summarization, and contradiction detection (GitHub)

Research

• Dynamic Gaussian Marbles for Novel View Synthesis of Casual Monocular Videos (SIGGRAPH Asia 2024, project page)

• Flash Cache: Reducing Bias in Radiance Cache Based Inverse Rendering (ECCV 2024, project page)

• StereoCrafter: Diffusion-based Generation of Long and High-fidelity Stereoscopic 3D from Monocular Videos (project page)

• Application of 3D Gaussian Splatting for Cinematic Anatomy on Consumer Class Devices (project page)

• GigaGS: Scaling up Planar-Based 3D Gaussians for Large Scene Surface Reconstruction (project page)

• Self-Evolving Depth-Supervised 3D Gaussian Splatting from Rendered Stereo Pairs (BMVC 2024, arXiv)

Misc

• It's not at all obvious to me that findings about how LLMs perform at coming up with ideas about how to prompt LLMs say anything about how LLMs perform at coming up with ideas about other things. I think this is a big problem for this whole field of research. (X)