ComfyUI-Ovi
Integrates Character.AI's Ovi joint video and audio generation engine into node-based ComfyUI workflows.
Generate synchronized video frames and 16 kHz audio tracks using precision-controlled Ovi nodes.
ComfyUI-Ovi brings Character.AI's dual video and audio generator directly into ComfyUI pipelines. Multimedia localizers and video content creators can generate synchronized speech and video prompts directly from custom workflow nodes. It suits team members building automated video translation or AI dubbing pipelines within ComfyUI.
What it is
This node package integrates Character.AI's 11B Ovi model into ComfyUI to output synchronized video frames and 16 kHz audio latents. It provides dedicated nodes for engine loading, precision switching, attention backend selection, and latent decoding. The tool automatically fetches core assets while allowing shared use of existing Wan 2.2 VAE and UMT5 text encoders.
Who it's for
It suits multimedia localizers, AI video creators, and technical translators building automated video dubbing or generation workflows in ComfyUI. It is not designed for non-technical users who lack an existing ComfyUI environment or high-end NVIDIA hardware. If you do not have at least 16 GB of VRAM, this tool will not run.
Key features
Self-bootstrapping model loader
Downloads required MMAudio assets and Ovi fusion weights directly into your ComfyUI directories.
FP8 and BF16 precision toggles
Lets you switch between full precision and a 16-24 GB VRAM mode directly from node dropdowns.
Attention backend selector
Allows runtime switching between FlashAttention, SDPA, Sage, and native attention drivers.
Component reuse support
Connects to pre-installed Wan 2.2 VAE and UMT5 text encoders without duplicating model files on disk.
Per-node multi-GPU targeting
Detects available GPUs and lets you assign specific workflow nodes to target graphics cards.
Up and running in 3 steps
Our honest verdict
What's great
- Generates synchronized video latents and 16 kHz audio in a single pass
- Reuses existing Wan 2.2 VAE and UMT5 text encoder files to save storage space
- Supports runtime switching between FP8 and BF16 precision modes
- Includes per-node device selection for multi-GPU hardware setups
Where it falls short
- Requires a minimum of 16 GB VRAM and a modern CUDA 12.x stack to execute
- No explicit open-source license provided in the repository metadata
- High VRAM retention after generation runs requires manual model unloading
- Requires setting up and maintaining a working ComfyUI environment
Pricing & license
The custom nodes are free to download and use. The primary financial cost comes from operating high-end GPU infrastructure capable of processing 11B models.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
What VRAM is required to run ComfyUI-Ovi?
You need 16 to 24 GB of VRAM using FP8 precision with CPU offload, or 32 GB+ VRAM for native BF16 mode.
Does ComfyUI-Ovi generate audio alongside video?
Yes, it produces synchronized video frames together with 16 kHz audio latents.
Can I use existing Wan 2.2 model files?
Yes, the Ovi Wan Component Loader node lets you point to your existing Wan 2.2 VAE and UMT5 text encoders.
