Skip to main content
ATI (Any Trajectory Instruction) is a controllable video generation framework proposed by the ByteDance team. ATI is implemented based on Wan2.1 and supports unified control of objects, local regions, and camera motion in videos through arbitrary trajectory instructions.

Project Repository

ATI source code and documentation on GitHub

Online Trajectory Editing Tool

Visual tool to draw and edit motion trajectories on images

​
Key Features

  • Unified Motion Control: Supports trajectory control for multiple motion types including objects, local regions, and camera movements.
  • Interactive Trajectory Editor: Visual tool that allows users to freely draw and edit motion trajectories on images.
  • Wan2.1 Compatible: Based on the official Wan2.1 implementation, compatible with environments and model structures.
  • Rich Visualization Tools: Supports visualization of input trajectories, output videos, and trajectory overlays.

​
WAN ATI Trajectory Control Workflow Example

Make sure your ComfyUI is updated.Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated.If nodes are missing when loading a workflow, possible reasons:
  1. You are not using the latest ComfyUI version (Nightly version)
  2. Some nodes failed to import at startup

​
1. Workflow Download

Update your ComfyUI to the latest version, then download and drag the workflow file into ComfyUI, or find “Wan2.1 ATI” in the Template Library under Workflow → Browse Templates → Video. Wan2.1 ATI Workflow Preview

Run on Comfy Cloud

Open in Comfy Cloud

Download Workflow

Download JSON or search “Wan2.1 ATI” in Template Library
We will use the following image as input:

Input Image: video_wan_ati_input_image.jpg

Download the default input image, or use your own image as the starting frame.

​
2. Manual Model Installation

All models involved in this guide can be found here. Diffusion Model : Choose one version:

Diffusion Model: Wan2_1-I2V-ATI-14B_fp8_e4m3fn.safetensors

Wan2.1 I2V ATI 14B diffusion model (fp8 precision). Place in ComfyUI/models/diffusion_models/
Text Encoders : Choose one version:

Text Encoder: umt5_xxl_fp16.safetensors

Full precision text encoder (higher quality, larger size). Place in ComfyUI/models/text_encoders/

Text Encoder: umt5_xxl_fp8_e4m3fn_scaled.safetensors

FP8 text encoder (lower VRAM usage). Place in ComfyUI/models/text_encoders/
VAE

VAE: wan_2.1_vae.safetensors

Wan2.1 VAE model. Place in ComfyUI/models/vae/
CLIP Vision

CLIP Vision: clip_vision_h.safetensors

CLIP Vision model for processing reference images. Place in ComfyUI/models/clip_vision/
File Storage Location

​
3. Complete the Workflow Execution Step by Step

Workflow step diagram Please follow the numbered steps in the image to ensure smooth execution of the corresponding workflow.
  1. Ensure the Load Diffusion Model node has loaded the Wan2_1-I2V-ATI-14B_fp8_e4m3fn.safetensors model
  2. Ensure the Load CLIP node has loaded the umt5_xxl_fp8_e4m3fn_scaled.safetensors model
  3. Ensure the Load VAE node has loaded the wan_2.1_vae.safetensors model
  4. Ensure the Load CLIP Vision node has loaded the clip_vision_h.safetensors model
  5. Upload the provided input image in the Load Image node
  6. Trajectory editing: Currently there is no corresponding trajectory editor in ComfyUI yet. You can use the following link to complete trajectory editing.
  7. If you need to modify the prompts (positive and negative), please make changes in the CLIP Text Encoder node numbered 5
  8. Click the Run button, or use the shortcut Ctrl(cmd) + Enter to execute video generation