A generated clip ends just as the scene becomes useful. Can an extension carry it forward without losing the scene? We tested that with a red paper boat, teal water and a pale stone in the background.
The capability
Google announced expanded access to AI video creation in Vids on 23 September 2026. Scene extension is one of the creative controls described in its official announcement. Our experiment examines one output, rather than treating that announcement as proof of reliability.
What we actually did
We generated a ten-second portrait clip in Google Vids, then used Extend once to request ten more seconds. The instructions asked for the same red folded-paper boat, a locked camera, continuous left-to-right movement, stable lighting and the pale stone in the upper-right background. We also asked for the entire boat to remain visible.
The interface identified the model as Omni. We did not independently verify the exact underlying model version. This account offered 720p generation; 1080p showed an upgrade requirement, despite the broader HD availability described in Google's announcement. We used the included access without upgrading. Our final 1080-by-1920 REVL video does not turn that source into native 1080p detail.
What the output shows
The red paper boat remains recognizable across the continuation, with the stone and warm lighting still present. Around the ten-second boundary, the scene reads as a continuation in our visual inspection. We added no crossfade or retouching to hide the join.
But the instruction to keep the whole boat visible did not hold. Near the end, the boat reaches and crosses the right edge of the frame. The base clip also begins with part of the boat outside the left edge. This is a useful distinction: visual continuity and full instruction compliance are different things.
What this test cannot prove
We generated the base twice because the first preview was lost during a browser download problem. The second used the identical prompt and was not selected through visual-quality retries. We made one extension request and kept that result. This is one scene, not a success-rate benchmark or a promise that people, products or more complicated motion will remain consistent.
The footage is entirely AI-generated. The article and graphics are original REVL work; narration is synthetic. Video excerpts play at normal speed. Carousel comparisons are explicitly labelled frames. The original full result is preserved with the production record.
A practical review checklist
Check the join: does the motion continue? Check the subject: do important details stay recognizable? Check the framing: does the longer shot still follow the brief? A convincing continuation can still need editing before it is a finished shot.
Test and official source checked 30 September 2026. Availability may differ by account. Follow REVL for AI, clearly explained.
AI, clearly explained.
Follow on Instagram · Follow on X
Explore REVL AI