First and last frame: controlling where an AI shot starts and ends

How to plan first and last frames for AI video so shots start and end exactly where you want: reveals, transformations, and shots that chain together.

October 14, 2026, 4 min read

The short answer

First and last frame control means giving the video model two images, one for where the shot begins and one for where it ends, and letting it work out the motion in between. It is the closest thing AI video has to keyframes, and it is how you get reveals, transformations and shots that cut together cleanly.

The trick is that the two frames must be compatible. Same lens feel, same light, same character, same world. The model fills the gap between them; it cannot bridge two images that look like they came from different films.

When to use it

Reveals: start on a close detail and end on the wide view, or start in darkness and end fully lit.
Transformations: an empty room becomes a furnished one, a bare tree comes into bloom, a sketch becomes a finished product.
Precise endings: you need the shot to end on a specific composition, for example with space for a title.
Chaining shots: the last frame of shot one becomes the first frame of shot two, so a longer sequence flows.
Product moments: a closed box ends open with the product sitting inside.
Match cuts: the end of one shot and the start of the next share a shape, so the cut feels designed.

Not every model accepts an end frame. Check the options shown for the model you pick before you plan a sequence around it. When a model does not support it, you can still plan with first frames only and let the cut do the work.

How to design the two frames

Make both frames in the same session, with the same image model, the same aspect ratio and the same description of the character and setting. Change only the thing that should change.

Keep the camera height and lens feel the same unless the camera is meant to move. If the end frame is a closer framing, that tells the model to push in.
Keep the light direction the same. If the key light is on the left in frame one, it stays on the left in frame two.
Keep the character's outfit and features identical. Use the same reference images for both.
Make the change readable. A small change produces a subtle shot; a huge change produces a morph. Decide which you want.
Avoid swapping positions of many objects. The model has to invent paths for each one, and that is where things melt.

Before and after: a mismatched pair

Before: frame one is a bright daytime photo of a café exterior shot from across the street. Frame two is a moody close-up of a latte on a table inside, lit by a warm lamp at night. Prompt: "Camera moves into the café."

Why it fails: the two frames disagree on time of day, light direction, color and space. The model has to invent a door, an interior, a walk to the table and a change from day to night in a few seconds. You get a dissolve-like morph instead of a camera move.

After: frame one is the café exterior at dusk, warm windows glowing. Frame two is the same café seen through the window from slightly closer, the latte on the windowsill table in the foreground, same dusk light. Prompt: "Slow push-in from across the street toward the café window, ending on the latte on the windowsill table. Pedestrians pass in the foreground. Warm window glow, dusk."

Try this prompt in Create Video

Now the frames share a time, a light and a space, and the change is just distance. The model can make that a real push-in.

A worked example: a three-shot chain

A short opener for a bakery, built from four images:

Image A: wide shot, the bakery street at dawn, shutters down, cool blue light.
Image B: same angle, shutters up, warm light spilling out, a baker in an apron placing a sign outside.
Image C: close-up of the baker's hands setting a tray of croissants on the counter, warm light.
Image D: same counter, slightly wider, the tray full and a customer's hand reaching in from the right.

Shot 1 runs A to B: "Morning arrives, the shutters roll up and warm light spills onto the street as the baker steps out with a sign. Camera locked off." Shot 2 starts on C with no end frame: "The baker sets down the tray, steam rising from the croissants. Slow push-in." Shot 3 runs C to D: "Camera eases back as a customer reaches for a croissant." Shots 2 and 3 share frame C, so the cut between them is invisible.

Try this prompt in Create Video

In ClickPop, storyboard mode in Create Video draws stills first, which is a natural way to produce these frames and approve them before rendering. Film Studio lets you build the shots one by one with set camera and lens choices and then cuts them into one graded film.

Prompts that work with keyframes

With both frames fixed, the prompt only needs to describe the journey.

"Smooth slow transition, camera stays locked off, the flowers open gradually." Try it
"Camera pushes in steadily, no cuts, the subject stays centered."
"The lights come up slowly from dark to full brightness, dust drifting in the air."
"The box lid lifts and slides away to reveal the watch inside, camera holds still."

Mistakes to avoid

Two frames from different sources. Generate them together, from the same description and references.
Too much change for the clip length. Long journeys need longer clips or extra keyframes in between.
Different aspect ratios. Both frames must match the output shape.
Expecting a cut. The model makes continuous motion. If you want a cut, make two shots.

Questions

What is first and last frame in AI video?

It is a way of giving the model a start image and an end image so it generates the motion between them. It gives you control over exactly how a shot opens and closes.

Why does my first and last frame video look like a morph?

The two frames are too different in light, space or composition, so the model blends them instead of moving through a real scene. Make both frames from the same setup and change only one thing.

Do all AI video models support an end frame?

No. Check the options shown for the model you choose. Where it is not supported, plan your sequence with first frames and cuts instead.