gen‑ai.news
← Back
Multimodal

ByteDance Releases Lance, a 3B-Parameter Unified Model for Image and Video Generation and Editing

ByteDance Releases Lance, a 3B-Parameter Unified Model for Image and Video Generation and Editing

Lance is a new open-source model from ByteDance's Intelligent Creation Lab that combines image understanding, image generation, video understanding, video generation, and editing into a single architecture using only 3 billion activated parameters. The goal is to replace the common practice of stitching together task-specific models with a single system that shares representations across modalities.

The unified design has practical implications beyond parameter efficiency. When a model is trained jointly on understanding and generation tasks across both image and video, it can draw on visual comprehension when generating - for example, applying knowledge of what a scene contains when editing only part of it. Separate models for each task lack that shared context and often produce edits that are inconsistent with the rest of the frame.

At 3B activated parameters, Lance sits in a range that makes it feasible to run on research hardware or reasonably sized cloud instances, which matters for an open-source release. ByteDance has made both code and weights available, allowing external researchers and developers to fine-tune or build on the model without going through an API.

The release arrives as several labs are pursuing similar unified architectures. The value of any single unified model ultimately depends on whether the joint training actually improves task performance rather than just reducing model count, and independent benchmarking of Lance's generation quality relative to specialised models will be the real test. ByteDance has not yet published detailed benchmark comparisons against task-specific alternatives.

Enjoy this story? Get the next one in your inbox.

Twice a week: the most important stories in generative image and video AI, distilled into a 2-minute read.

Free. Unsubscribe any time. No spam, ever.

Your next read

No image
Multimodal

Fine-tune video and image models at scale with NVIDIA NeMo Automodel and 🤗 Diffusers

NVIDIA and Hugging Face have joined forces to bring large-scale fine-tuning of image and video diffusion models into the NeMo Automodel framework, integrated with the Diffusers library. The collaboration aims to make distributed training more accessible for teams working with models that would otherwise be difficult to fine-tune on limited hardware. The result is a more streamlined path from a pretrained diffusion model to a customized one, without requiring deep infrastructure expertise.

No image
Multimodal

Thinking Machines Lab Releases Inkling: A 975B-Parameter Open-Weights Multimodal MoE With 41B Active Parameters And Controllable Thinking Effort

Thinking Machines Lab has released Inkling, a 975B-parameter open-weights multimodal model built on a Mixture-of-Experts architecture that keeps only 41B parameters active at any given time. Licensed under Apache 2.0, it accepts text, image, and audio inputs and offers a 1M-token context window. Rather than competing for top benchmark rankings, the model is positioned as a customizable base with adjustable reasoning depth.