AEmotionStudio/Video-Depth-Anything-Large

Video Depth Anything — Large

Mirror of depth-anything/Video-Depth-Anything-Large for use with ComfyUI-FFMPEGA.

What is Video Depth Anything?

Video Depth Anything is a state-of-the-art model for temporally consistent monocular depth estimation in videos. It extends Depth Anything V2 with temporal modules for smooth, flicker-free depth maps across video frames.

Key features:

• Temporal consistency — smooth depth maps without frame-to-frame flickering
• Multiple encoder sizes — Small (335.3M), Base, and Large variants
• Apache 2.0 license — fully open source
• Colormap output — supports multiple colormap visualizations (inferno, magma, plasma, etc.)

Files

model.safetensors
config.json

Usage

With ComfyUI-FFMPEGA (recommended)

1. Set no_llm_mode to video_depth on the FFMPEG Agent node
2. Select encoder size (small, base, large) under Advanced Options
3. Choose colormap for visualization
4. The model auto-downloads on first use

Manual download

huggingface-cli download AEmotionStudio/Video-Depth-Anything-Large --local-dir ./video_depth_anything

Available Sizes

Variant	Parameters	Size	Speed
Small	24.8M	~102 MB	Fastest
Base	97.5M	~390 MB	Balanced
Large	335.3M	~670 MB	Best quality

License

Apache 2.0 — see the upstream repository for full license terms.

Credits

• Original model by: Depth Anything team
• Paper: "Video Depth Anything: Consistent Depth Estimation for Super-Long Videos"
• Upstream HuggingFace: depth-anything/Video-Depth-Anything-Large
• Redistributed by: Æmotion Studio for use with ComfyUI-FFMPEGA