First and last frame generation (also called interpolation or start/end frames) lets you upload two images: where the clip starts and where it ends. The model generates the motion between them. It is the most reliable way to control where a shot lands, which makes it popular for product reveals, transformations and scene transitions.
How it works
A normal image-to-video generation uses your image as the opening frame and invents everything after it. With first and last frames, the model has a fixed destination, so it plans the motion to arrive at your second image. Your text prompt describes how to get there: the action, camera movement and sound.
Which models support first and last frames?
- Google Veo 3.1: supports interpolation; the last frame must be combined with a first image, per Google's documentation. Clips are 4–8 seconds with audio.
- Gemini Omni Flash: supports first and last frame interpolation and uses
<FIRST_FRAME>and<LAST_FRAME>tags in prompts. - Kling 4.0: supports first and last frames plus up to 10 keyframes for multi-step transitions.
- MiniMax H3: its first/last-frame mode accepts zero, one or two images (start, end or both), per its model card.
Availability varies by platform. In the TextVidAI generator, check the options for each model.
Step by step
- Prepare two images with the same aspect ratio and similar lighting.
- Upload the start frame and the end frame.
- Describe the transition in the prompt: what moves, how the camera travels, and the sound.
- Pick a length that fits the change. Small changes need 4–5 seconds; big transformations need more time or a middle keyframe.
- Generate, review, then adjust one thing at a time.
Prompt examples
- Product reveal: Start: closed gift box on a table. End: the watch from the box on a velvet stand. The lid lifts, the camera pushes in and the watch rises into the light. Soft click and a subtle chime.
- Day to night: Start: city street at noon. End: same street at night with neon signs. Time-lapse transition, clouds race, lights switch on one by one, static camera.
- Before/after: Start: empty living room. End: furnished living room. Furniture slides into place smoothly, camera slowly orbits right.
- Character move: Start: woman sitting on a bench. End: same woman standing at the railing looking at the sea. She stands up, walks three steps and leans on the railing. Waves and seagulls.
Tips for smooth transitions
- Match the camera position. If the angle changes a lot between frames, describe the camera move that connects them.
- Keep the subject identical. Use the same character, clothing and colors in both frames.
- Avoid impossible jumps. If the two frames have nothing in common, add a middle keyframe or split into two clips.
- Same lighting and color temperature prevent flicker at the end of the clip.
- Create the frames with an image model first if you don't have photos, then animate between them.
Related guides
- How to turn a photo into a video with AI
- How to keep characters consistent in AI video
- Kling 4.0 prompt guide (keyframes)
FAQ
What is first and last frame in AI video?
It is a mode where you upload a start image and an end image and the AI generates the motion between them.
Is it the same as keyframes?
It is the simplest form of keyframing: two keyframes. Models like Kling 4.0 support more keyframes in one clip.
Why does my transition look like a fade?
The frames are too different. Make them more similar, describe the motion clearly, or add a middle keyframe.
Sources
- Google: Veo 3.1 documentation
- Google: Gemini Omni Flash documentation
- Kling: Introducing Kling 4.0
- MiniMax H3 model card
Last updated: October 9, 2026.