Load Video, Set Resolution & Shotname

After uploading your video, you can adjust the resolution in the Resize Image node—though I recommend keeping it at 720p.

Assign SET SHOTNUMBER a unique number or name for each clip you want to prepare. This saves the outputs in separate folders, making it easy to track the start frame, end frame, and mask video.

Start Frame & End Frame—Manual Bridge

Preview Bridge isn't currently working, so we need an alternative solution for inpainting.

Until it's restored, this manual workaround works well:

1. Each Preview Image will grab the first and last frames of the video.

2. Right-click on the Preview Image and select Copy Image.

3. Select the Load Image node and press Ctrl + V to paste the image.

4. You can then inpaint on it—the mask even saves between sessions!

Startframe & Endframe - Qwen Remove

Here we use Qwen to remove the stick. A simple prompt like "remove the wooden stick" works well. If results are inconsistent, try a different seed or expand the mask area.

Sam3 Mask Subject

This section automatically creates a mask for your character. If sam3 is too difficult to install, we've provided a sam2 workflow below this section.

Use Pick Track Frame to select a frame where your character is most visible. Load your image by pressing ▶︎, then left-click to select your character. Right-click to set negative points. We use this to exclude the stick from the mask.

Now you can run the workflow!

Time2Move Workflow

🎨 Workflow Sections

⬜ Input / Output / Model Loaders

🟩 Setup Video

🟪 Sampler

🟨 Important Notes

Load Input

Load everything generated in the Preparation Workflow here. You can upload both the start and end frames, upload only the start frame, or upload only the end frame—all three options work. However, both the input video and mask video are always required.

Setup Video

Here you can adjust settings for this workflow. The Resolution switch lets you choose between 576p at 1 and 720p at 2. Set it to 3 to use the Custom Width and Custom Height attributes. If you've uploaded only a start frame or only an end frame, disable the one you haven't loaded.

Sampler

Now you can generate your video. We've found that both very precise and very short prompts work well in different circumstances. Start with a short prompt like "A plush dinosaur jumps," then add more detail if needed.

The WanVideo Add TTMLatents is currently set to end_step 2. You can set it to 3 if you want the result to follow the input video more strictly, though our results didn't look as good with that setting.