I’d like to propose a feature request for the Gemini / Veo video generation API that I believe would unlock a very compelling creative and commercial use case:
*Support for “last-frame-only” conditioning, where a user can provide *only a final frame image (without a required first frame), and the model generates all preceding frames guided by text prompts—ending exactly on the provided last frame.
Why this matters
1) Guaranteed product consistency
In advertising and product storytelling, visual consistency is critical. Allowing only the last frame to be fixed ensures the product appears exactly once and exactly as provided, with 100% fidelity. This avoids partial hallucinations, distortions, or early reveals while keeping the product pristine at the final moment.
2) Strong audience hook via reveal dynamics
This enables highly engaging “reveal-style” videos:
- The model can generate dynamic, creative, cinematic opening frames (camera movement, abstraction, motion, mystery).
- The sequence then bridges into a precise final product frame.
- This creates a powerful hook → payoff pattern that is extremely effective for short-form video (ads, social, product launches).
3) Clean mental model for creators and developers
Conceptually:
- Text prompt controls the story and motion.
- Last frame image guarantees the final visual outcome.
This separation is intuitive and maps well to real-world creative workflows.
Technical note
I understand Veo currently supports:
- Text-to-video
- Image-to-video (first frame)
- First + last frame interpolation
This request is specifically for enabling last_frame usage without a required first frame—allowing the model to freely generate the starting frames while converging to a fixed final image.
If this is already technically feasible but restricted by the current API contract, I’d love to understand the constraints. Otherwise, I believe this would be a highly differentiated capability for Veo compared to other video models.