I’ve been experimenting with a fun idea: continuous video generation.

You give it a scene prompt (and optionally a reference image), it generates a clip — then it can keep extending the same video by automatically coming up with the next scene and appending it.

What I Built

A workflow that:

• Takes a prompt (+ optional reference image)
• Generates the first clip with Sora2
• Repeats a loop to extend the video indefinitely

How It Extends the Video (High Level)

Each cycle looks like this:

1. Grab the current full video
2. Extract the last frame
3. Generate the next scene prompt from the last frame + previous context
4. Generate the next clip and stitch it onto the end

That last-frame handoff is what keeps the continuity surprisingly stable.

Output Example

Here is a continuous video generated by this workflow, merging two separate clip generations:

Tutorial (Recorded, Not Posted Yet)

I recorded a full tutorial walking through the workflow, but it’s not published online yet.

What’s Next

The next improvement I want is better pacing logic (when to hold a shot vs. introduce a new beat) while keeping the workflow simple.