First and last frame AI video: prepare the pair before the motion.
Use first and last frames when a supported video mode needs to connect a specific opening image to a specific ending image. Choose two views that can belong to the same shot, assign them to the actual start and end controls, then describe the movement between them. Two uploaded images alone do not tell every tool which one is the ending.
1. Check the mode before preparing the second image
| Input or workflow | What you are planning | Check before using it |
|---|---|---|
| A starting image | Motion beginning from one image. | Whether an explicit ending-image input is available. |
| Explicit start and end images | One generated shot connecting the two endpoints. | The selected model, mode, and assignment of each file. |
| Reference images | Visual guidance, such as an object or character reference. | Whether the interface actually assigns any reference to a time in the clip. |
| Two clips joined in an editor | A cut or an editing transition. | Whether you need continuous generated motion at all. |
One documented example is Google Cloud's first and last frame workflow. Its Veo instructions use Image-to-video in Media Studio, with separate Start and optional End uploads. The page lists Veo 3.1 generation models, including veo-3.1-generate-001 and veo-3.1-fast-generate-001. This was checked October 15, 2026; it does not establish support in every app that uses the Veo name. Confirm the controls in your own route before requesting a generation.
2. Make the two images agree on what stays the same
Put the images side by side at the intended delivery crop. First identify the change you want to see. Then look for differences that have nothing to do with that change. A box opening is one task; a box opening while the room, viewpoint, and object all change is a much less specific brief.
- Subject: match the object's shape, markings, number of parts, and important proportions.
- Composition: compare its position and size within the crop. Keep the viewpoint fixed if the intended shot uses a fixed camera.
- Scene: check the background, surface, light direction, and shadows. Decide which differences are intentional.
- Space for motion: leave room for the lid, hand, or moving subject throughout its path, not just at the endpoints.
- Files: prepare both images in a compatible crop and the formats accepted by the selected mode. Check its current size and resolution requirements.
These are preparation checks, not guarantees that the model will preserve every feature. Use the aspect-ratio guide when the crop changes the composition. If the sequence contains a person, also compare the character continuity details before treating two images as the same person in the same shot.
If one image shows a box on a table and the other shows a person in a different room, decide whether you want a visible transformation or simply the next shot. Use a cut when the story needs a new scene; do not ask interpolation to hide that decision.
3. Example: one box, one hinge, one change
The following original side-view sketches are planning diagrams, not generated output or ready-to-upload source frames. Prepare your own actual endpoint images. The intended action is a plain green box's attached lid lifting from closed to partly open while the camera stays fixed.
Start image — lid closed
Keep the whole box and the hinge visible. Leave empty space above it for the lid's movement.
End image — the same lid partly open
Keep the body, hinge position, surface, crop, and viewpoint consistent. Change the lid's angle rather than replacing the box or adding a new scene.
Motion brief: The green box stays in place as its attached lid lifts slowly on the rear hinge and comes to rest partly open. The camera holds a fixed side view, with steady light on the tabletop.
This is an untested prompt example. It describes the path between the images; it does not make a file an endpoint. Upload the closed image to the selected mode's start control and the open image to its end control. Do not rely on file names or phrases such as “image two” unless that interface documents them as supported syntax.
Copy this endpoint brief before generating
Provider / model / mode: Explicit start and end inputs confirmed: Start image file: End image file: Delivery crop: What stays the same: One visible change: Subject's path through the shot: Camera behavior: Motion prompt: Ending needed for the edit: Observed result / next revision:
For the box example, “what stays the same” includes the box body and hinge; “one visible change” is the lid angle. If the real frame pair already differs in the box's width or the camera's height, correct that pair before adding a longer prompt.
4. Inspect the opening, the path, and the ending
Watch the whole take, then inspect frames near the beginning and end. Check the middle separately: two plausible endpoints can still be connected by an unwanted movement. In the box example, the lid should remain attached to the same hinge instead of sliding away and returning.
| Observed result | Check first | Next action |
|---|---|---|
| The ending image seems ignored. | Model, mode, and the actual end-image assignment. | Confirm the supported input and file before rewriting the prompt. |
| The object changes shape during the motion. | Differences in geometry or viewpoint between the images. | Prepare a closer-matching pair; simplify the intended change. |
| The camera moves unexpectedly. | Whether the endpoints imply different framing. | Align the composition or deliberately plan a camera move. |
| The middle rushes through several actions. | How much change the pair and prompt request. | Reduce the action or split the idea into separately planned shots. |
| The final frame must match an approved still exactly. | The accepted output against that actual still. | Finish the hold or cut in an editor and inspect the handoff. |
If a single transition has become several story beats, use the storyboard and shot-card template to separate them. For a planned camera move, name it clearly with the camera movement guide. Save the accepted take, the two input files, and the exact settings so the next revision addresses an observed problem.
Common first and last frame questions
Can I upload two unrelated images?
You can plan a deliberate transformation, but two unrelated images do not define a natural continuous shot. Decide whether to prepare a more compatible pair, show the transformation explicitly, or use two shots and a cut.
Will using the same image twice make a seamless loop?
Matching endpoint images do not establish matching movement, speed, or sound across a loop boundary. Review the last-to-first join in an editor. Do not assume the repeated image makes the whole clip seamless.
Does an end-frame input guarantee an exact final still?
Check the actual output and the selected mode's documented behavior. If your edit needs an exact approved still, plan the hold or cut to that still in your editor and review the join rather than promising a match from the prompt.
Can Shotmelo generate the transition?
The current preview lets you edit a prompt, preview a reference locally, save a draft, and export text. AI image/video rendering, accounts, and payments are not connected. Endpoint uploads and video generation happen in the tool you choose.
Once the pair is ready, draft and save the motion text in the Shotmelo planning workspace. Keep the actual endpoint file assignments in your brief and in the generation tool's supported controls.