Image to Video AI: Control the Shot with Start and Last Frames

Key takeaways
- Put an image in the start frame and the video begins from that exact scene.
- Add a last frame too and the model animates the change between the two images.
- The image sets the aspect ratio: a vertical image gives a vertical video.
- In FastMetal Studio, seven video models take a start frame and four also take a last frame.
- Settling the composition in a still first gets you to the shot you want faster than rerolling text-to-video.
Text-to-video changes the framing and the subject's look on every run. If you already have an image you like, make it the first frame and the model keeps that composition while it adds motion. Here is how it works in FastMetal Studio.
Start frames and last frames
The start frame is the first frame of the clip; the last frame is the final one.
- Start frame only: the clip begins on your image and the model continues it from the prompt. Good for animating a product photo or an illustration.
- Start and last frame: the model generates the motion between two images. Good for before-and-after changes: a shirt changing color, day turning to night, a box opening.
Supported video models
As of November 2026, these Studio models accept frames:
| Model | Lab | Start frame | Last frame |
|---|---|---|---|
| Seedance 2.0 Fast | ByteDance | Yes | Yes |
| Seedance 2.5 | ByteDance | Yes | Yes |
| Wan 3.0 | Alibaba | Yes | Yes |
| Luma Ray 3.2 | Luma | Yes | Yes |
| Kling 3.0 Pro | Kuaishou | Yes | No |
| Hailuo 2.3 | MiniMax | Yes | No |
| HappyHorse 1.1 | Alibaba | Yes | No |
Luma Ray 3.2 does not offer 10 seconds when a frame is used. Seedance 2.5 Flash is text-to-video only.
Step by step in Studio
1. Get the first image
Use your own photo or make one in Studio's Create image tool. If you generate it, use the aspect ratio you want for the video, because the video inherits it: vertical for Reels, TikTok and Shorts, horizontal for YouTube.
Images are billed per image at a lower rate than a video clip (see pricing). Work out the framing, light and the subject's look here, over a few stills.
2. Open Create video and pick a model
Choose a model from the table that supports a start frame, or one that supports a last frame if you need both.
3. Choose the frames
Use "Choose start frame" to pick an image from your library or upload one. For a last frame, use "Choose last frame".
4. Describe the motion
The prompt does not need to describe what is already in the image. Describe what moves:
The camera slowly orbits to the right. Steam rises and the tree outside the window sways in the wind.
With a person in the frame, one or two sentences on what they do ("slowly turns and smiles") work well.
5. Pick the quality and generate
The image fixes the aspect ratio, so you choose the quality instead. The price is on the Generate button before you click.
Picking images that work
- One clear subject. Images with many small figures tend to break up once they move.
- Shoot or generate start and last frames in the same place and light. A big gap between them makes the motion in between look unnatural.
- Text in an image can warp as it moves. Add logos and captions in an editor after generation.
Common pitfalls
- Your images are sent to the provider of the video model. Do not use anything that cannot leave your company.
- Once a frame is set, the aspect ratio picker disappears; the image's ratio is used.
- For a last frame, pick a model that supports one (see the table).
FAQ
Is image to video AI free?
Video on FastMetal is paid per clip, from a prepaid balance with no monthly fee. Rates are on the pricing page.
Can I use my own photos?
Yes. Upload them from the start frame picker. If a person appears, use photos they have agreed to.
Is the first frame of the video identical to my image?
The clip starts on the same composition, but some models shift colors or small details slightly.
Can I set frames through the API?
Start and last frames are available in Studio today. The API steps for video generation are in the video generation docs.
Bottom line
Settle the composition in a still, then describe only the motion in the video prompt. To see how each model renders the same prompt, read our comparison of five AI video models. Create a free account to try it in Studio.