Most current models struggle to understand when a specific event should happen, how long it should stay on screen, and what order different actions should appear in. Researchers introduced Wan 2.2 with Prompt Relay, a simple yet powerful framework designed for fine-grained temporal control in multi-event video generation. The biggest promise of this method is that it works entirely at inference time. That means it does not require retraining the model, changing the architecture, or adding extra computational cost.
Instead, it acts as a plug-and-play enhancement that improves how the model handles multiple prompts across different time segments. With Prompt Relay, video generation becomes more structured and controllable. Each part of the video can focus on its assigned semantic concept without interfering with other moments in the timeline.
 |
| Prompt Relay Explanation |
Wan 2.2 with Prompt Relay provides a clean and efficient solution for generating multi-event videos with better temporal consistency. The framework works by dividing a video into temporal segments and assigning a dedicated prompt to each segment. During generation, the modified cross-attention mechanism ensures that frames within a segment primarily focus on their corresponding prompt instead of the entire paragraph.
Installation
1. First, you should do the ComfyUI installation. Update ComfyUI to its latest version from the Manager.
2. Install the Prompt Relay custom node from Kijai's repository or alternatively use the Manager to install it.
Clone the repository inside ComfyUI/custom_node folder, using following command:git clone https://github.com/kijai/ComfyUI-PromptRelay.git
3. Make sure you do the basic Wan2.2 setup while running the workflow. If not done yet, follow our Wan2.2 installation tutorial.
4. Same models are also listed here:
(a) Download Wan 2.2 (High and Low) model. Save this inside ComfyUI/models/diffusion_models folder.
(b) Use the Wan2.2 Remix model.Save this into ComfyUI/models/diffusion_models folder.
If using low VRAM, use the Wan 2.2 Remix GGUF instead and save this into ComfyUI/models/unet folder.
5. Restart and refresh the ComfyUI.
Workflow
1. Download workflow (Wan2.2_PromptRelay.json) from our Hugging face repository.
2. Drag and drop into ComfyUI. Install the missing (red outlined colored) nodes by selecting "Install missing custom nodes" option from the Manager.
3. Load all the necessary(Wan2.2 High-Low, text encoders, vae etc) models into their respective nodes. Set the basic KSampler settings.
4. Add the prompts into prompt box. Follow the instruction you need to do the prompting. For more details follow the Kijais Prompt relay repository.
5. Input image into Load image if using I2V pipeline.
6. Hit Run to start generation.
Wan 2.2 with Prompt Relay is a strong step toward more controllable and cinematic AI video generation. Instead of relying purely on larger models or more training data, this approach improves generation quality through smarter attention control. What makes this method especially valuable is its practicality.
Many advanced AI techniques demand retraining, additional hardware resources, or complicated architectural changes. Prompt Relay avoids all of that while still delivering noticeable improvements in temporal coherence and prompt accuracy.
