Skip to documentation
On this page

Open Source Model

Overview

Vimoo 1.0 is an open-source video generation model built on the MMDiT architecture. It supports text-to-video (T2V) and image-to-video (I2V), using prompts or reference images to create videos with control over subjects, actions, scenes, and camera movement. Its lightweight variant, Vimoo 1.0 - Lite, supports both tasks in a single model designed for efficient inference. See Vimoo 1.0 for architecture, variants, and capabilities.

Open-Source Resources

Models are published in the Vimoo-1.0 Collection on Hugging Face.

Model familyHugging FaceInference codeIntegrations & ToolsIncluded models
Vimoo 1.0Vimoo-1.0Vimoo-1.0DiffusersT2V, I2V
Vimoo-1.0-ComfyUIComfyUI-VimooComfyUIT2V, I2V and their Distilled versions
Vimoo 1.0 - LiteVimoo-1.0-LiteVimoo-1.0DiffusersTI2V
Vimoo-1.0-Lite-ComfyUIComfyUI-VimooComfyUITI2V and its Distilled version
Vimoo-1.0-Lite-MLXVimoo-1.0-MLXMLXTI2V Distilled on Apple Silicon

T2V generates video from text; I2V adds a starting image. TI2V supports both tasks in one model. Lite is the lightweight model; Distilled versions use fewer sampling steps for faster generation.

Run Locally

Choose a framework for your device. For running locally, we recommend Distilled or Lite TI2V Distilled models.

Generation Time by Device

For running locally, we recommend Vimoo 1.0 T2V Distilled / I2V Distilled, or Vimoo 1.0 - Lite TI2V Distilled.

Use the chart to compare the two text-to-video choices at 720p, 121 frames. It covers Vimoo-1.0 T2V Distilled and Vimoo-1.0-Lite TI2V Distilled; I2V timings are not shown. Times vary with generation settings and runtime environment.

Generation times by device for Vimoo-1.0 T2V Distilled and Vimoo-1.0-Lite TI2V Distilled, in minutes

NVIDIA timings use SageAttention and torch.compile, measured after warm-up; compilation and warm-up time are excluded. Mac timings use MLX.

Server Deployment