Today in Generative Media

Luma's AI video model; Real photo wins AI contest; Google's AI alphabet generator

News

• ‘We don’t need Sora anymore’: Luma’s new AI video generator Dream Machine slammed with traffic after debut (VentureBeat)

• Luma launches Dream Machine (Luma Labs) Some fun examples:

  • Image to Video within seconds. (X)

  • I am absolutely BLOWN AWAY (X)

  • Here's a variety of images I ran through the Luma Dream Machine (X)

  • A new cinematic AI piece, well... more like a short film. (X)

  • I took the opportunity to create the opening sequence for a new project I'd like to develop (X)

  • Last one before bed. Night all, sweet generations! (X)

• Photographer Disqualified From AI Image Contest After Winning With Real Photo (PetaPixel)

• I tried Google's new AI alphabet generator, and it's way more fun than it sounds (ZDNet)

• Adobe employees slam the company over AI controversy: 'Let's avoid becoming like IBM' (Business Insider)

• Forget DALL-E: Apple's new AI image generator runs on-device and works like magic (ZDNet)

• AI speech-to-text can hallucinate violent language (TechXplore)

Software

• Create your own personalized images by ranking image pairs on midjourney.com and use the personalization parameter on your images. (MidJourney)

• SD 3 is ready for download (X)

  • 1 Click Stable Diffusion 3 with ComfyUI (X)

  • SD3 on Replicate (X)

• On my 21st birthday, I built Render — the first AI-Powered Video game generator. It can create entire video games from text. (X)

• Audio input now live in Suno. You can upload your own sounds and use them as the base for any song you can imagine (X)

Research

• Image Neural Field Diffusion Models (CVPR 2024, project page)

• Separating the "Chirp" from the "Chat": Self-supervised Visual Grounding of Sound and Language (CVPR 2024, project page)

• Neural Gaffer: Relighting Any Object via Diffusion (project page)

• ID-to-3D: Expressive ID-guided 3D Heads via Score Distillation Sampling (project page)

• An Image is Worth 32 Tokens for Reconstruction and Generation (project page)

• BitsFusion: 1.99 bits Weight Quantization of Diffusion Model (project page)

• MultiEdits: Simultaneous Multi-Aspect Editing with Text-to-Image Diffusion Models (project page)

• MOFA-Video: Controllable Image Animation via Generative Motion Field Adaptions in Frozen Image-to-Video Diffusion Model (project page)

• 4Real Towards Photorealistic 4D Scene Generation via Video Diffusion Models (project page)

• AsyncDiff: Parallelizing Diffusion Models by Asynchronous Denoising (project page)

• MimicBrush: Zero-shot Image Editing with Reference Imitation (project page)

• Human 3Diffusion: Realistic Avatar Creation via Explicit 3D Consistent Diffusion Models (project page)

• What If We Recaption Billions of Web Images with LLaMA-3 ? (project page)

• Cinematic Gaussians: Real-Time HDR Radiance Fields with Depth of Field (arXiv)