Using AI-generated video clips as source material for Interactive Video?

Most of the effort in an Interactive Video isn't the interaction authoring — it's getting a clean, short clip with enough visual headroom to place an overlay question without covering the subject. A few things I keep coming back to when preparing source material:

• Keep clips short (roughly 10–20 s). Long clips make it hard to decide where an interaction should actually sit, and viewers lose the thread before the question appears.
• Leave headroom in the frame. Overlays are anchored, so a clip that fills the frame edge-to-edge forces you to cover the focal point.
• Prefer multi-shot clips with visible scene changes — they map naturally onto chapter-style interactions and give you obvious pause points.

For placeholder and B-roll footage, text-to-video tools are now good enough to use as stand-ins while the real shoot is still being arranged. I've been using Kling3AI for 4K multi-shot clips with native audio, which removes a lot of the back-and-forth when you just need something close to the final look before committing to a shoot.

Two things I'd like the community's take on:

• Is there a repeatable way to keep overlay positions consistent across clips with different aspect ratios, or do you end up re-tuning every video?
• Any pitfalls with how H5P handles higher-bitrate source files for viewers on slow connections?

Curious whether others are mixing generated footage into their H5P authoring pipeline, and how you handle the hand-off between placeholder and final media.