Beta v0.1, workspace is a little cluttered, but fully functional. (img2vid Effect - Person Detection zoom-outpaint v01)
Here's a simple video effect that you can integrate into your projects:
Automatically segment a person/object, mask them, and zoom them in/out to create a video snippet.
See https://civitai.com/posts/2101126 for example.
How to use:
(Green Boxes are your focus for selections)
(This workflow was built using an LCM checkpoint. If you are not, then change the KSampler to 30 steps)
Load an image
Select your checkpoint for outpainting, set any pos/neg prompt for outpaint
Load a person_yolov8 model or one that fits your subject
Select the object/person with the 'take start' value in 'Detect Person A'
Set 'Size / Frame' to 0
Optionally set multiplier, default is set to 10% per frame, so 10 frames = 200% size
Change the 'Save Image' file path to a folder of your choosing, this will generate individual images/frames so make it unique for each project.
Set 'Size / Frame' control_after to Increment
Queue Prompt 10 times (for 10 frames)
Set 'Size / Frame' control_after to Decrement
Queue Prompt 10 times
Run these images through a VHS combiner, example workflow is provided and integrates frame-rate increase/blend. (img2vid - VHS Combine with 2x frames)
You can play with the other values. I tried to label them logically.
Let me know your thoughts in the comments, and as always share your generations!