actionsampler-wigglegram

---

Wigglegrams were a thing from the 35mm analog film era that can now be revived with the new Actionsampler camera. It is a very cheaply made camera that eats 35mm film and exposes 4 separate images onto a normal 35mm film frame. In theory this allows for two interesting use-cases to take creative pictures.

• closeup still shots creating a "3D" effect
• short 1s videos at 4fps

3D effect	video
selfie with a torch light straight to my face (still kinda underexposed)	a train in Onomichi Japan, summer 2025

Naturally, the 4 images on the 35mm photo will be aligned to nothing. This pyhon script aims tries to auto-align and crop the images to then use with ffmpeg to create the short gifs.

the camera

---

How the script works

The alligner performs the following steps in order:

1. import scanned analog image
2. add black border around the image
3. crop the image into equal 1/4 slices
4. convert the slices to HSV space
5. use the first slice (top-left) as reference to compare against
6. pick a square (1/4 wide) on the 3 remaining slices
7. bruteforce for loop searches for the smallest L2-norm difference of a sliding square window on the reference slice
8. (overlay all squares onto the 1. slice to verify correct convergece)
9. transform the square coordinates back to slice coordinates
10. create a large black image
11. (overlay all images to verify correct coordinate transform)
12. save each of the 4 offset images

input image

Step. 7	Step. 10
squares correctly matched to reference image	all 4 images overlayed (with black border)

---

Create a GIF

ffmpeg -framerate 8 -pattern_type glob -i './*.jpg' -c:v libx264 -pix_fmt yuv420p face.mp4
ffmpeg -i face.mp4 face.gif

---

Limitations and known issues

Using this script is a semi-automatic process. For some images the auto alignment might fail (partially) and manual adjustment of (some of) the coordinates is required.

Since the alignment is a brute force method, it might take a few seconds to process each subframe. The search radius and the square size are parametrizable and have a significant impact on the processing time. Expect ~1min for the provided example image (1024x1536).

The script may fail if:

• the subject is not at the center of each frame
• very dynamic shots that capture a lot of movement
• very empty subjects (like a wall / sky)

The script works well on:

• still subjects (a face)
• "busy" landscapes

Other considerations:

• should the wigglegram center and stabilize the subject (still face) or the background (moving subject)!

---

Requirements

The script was created and used in spyder using the following modules:

• numpy
• opencv

conda install -c menpo opencv

Acknowledgements

@jyjblrd | 3D-Printed-Film-Video-Camera for the ffmpeg commands