AI BriefingStability AI Research news
The latest AI and developer news from Stability AI Research, with short summaries.
Visit website →Arbor introduces constraint meshes for controllable 3D asset generation
Arbor uses hull, avoidance, and touch regions as native 3D control interfaces to guide latent generation.
4Director Controls Video World Models with Rigid 3D Geometry
The model uses a new dataset of 20,774 clips and a Motion Adapter to ensure view-consistent video generation.
SemanTok: Predictable Semantic Tokens Enable Efficient Autoregressive Video Generation
A 201M-parameter SemanTok AR model matches or outperforms a VideoFlexTok AR model 3.4x its size.
Stable-Layers: Fine-Tuning an Image Layer Decomposition Model via Reinforcement Learning Using VLM Scores
Stable-Layers fine-tunes an image layer decomposition model without paired data through reinforcement learning that leverages feedback from a VLM.
OCTOPUS: Optimizing Transformer KV Cache via Optimal Squared-Error Quantization and Octahedral Parameterization
**OCTOPUS**, which maximizes transformer KV Cache compression performance by combining octahedral parameterization with joint quantization, has been released.
Introducing Stable Audio 3 and SAME (Semantically-Aligned Music Autoencoder)
Released Stable Audio 3, a fast latent diffusion model for variable-length audio generation and editing.
Foley Control aligns a frozen latent text-to-audio model to video
Foley Control attaches only a small video cross-attention to a frozen T2A model to learn video synchronization.
Stable Video Materials 3D generated from a single image
It generates consistent PBR materials and normals from a single image to support relighting.
ReSWD: Steady Like ReSTIR - Variance Reduction by Combining Reservoir Sampling and Sliced Wasserstein Distance
By combining Reservoir Sampling with SWD, it reduces high-variance estimation and unstable gradients.
Stable Cinemetrics: A Structural Taxonomy and Evaluation for Professional Video Generation
By dividing professional video generation into 4 control axes, it exposed the limitations of 10+ models.
Music and Artificial Intelligence: Artistic Trends
AI in music is expanding into a co-creation tool, medium, and performance format.
SD3.5-Flash: Distribution-Guided Distillation for Generative Flow
SD3.5-Flash is a lightweight distillation framework that produces high-quality images in just a few steps.
Stable Part Diffusion 4D: Generating Multi-View RGB and Kinematic Part Videos from a Single-View Input
SP4D generates RGB and kinematic part videos together from a single-view input.
Material Reconstruction and Blending in CLIP Space
By finding material embeddings in CLIP-space, it implements material blending and fine-grained attribute control.
Fast Text-to-Audio Generation via Adversarial Post-Training
ARC post-training generates 12 seconds of 44.1kHz stereo audio in 75ms on an H100.
Generating Animatable 3D Face Avatars from a Single Image
It generates 4D face avatars with multi-view consistency and expressions from just a single image.
SV4D 2.0: Enhancing Spatio-Temporal Consistency in Multi-View Video Diffusion for High-Quality 4D Generation
SV4D 2.0 is a Multi-View Video Diffusion model that generates high-quality dynamic 3D assets by enhancing spatio-temporal consistency.
High-Speed, High-Resolution Image Synthesis Using Latent Adversarial Diffusion Distillation
SD3-Turbo, which generates high-resolution images with just 4 sampling steps, was introduced through LADD technology.
Stable Virtual Camera: Multi-View Video Generation Using 3D Camera Control
Stable Virtual Camera is a general-purpose diffusion model that generates highly consistent multi-view videos through 3D camera control.
SPAR3D: Stable Point-Aware 3D Object Reconstruction Based on a Single Image
SPAR3D leverages a point diffusion model to reconstruct high-precision 3D objects from a single image in just 0.7 seconds.