# What is first and last frame in AI video?

URL: https://dreambench.ai/glossary/first-and-last-frame

First and last frame, also called start and end frame or FLF2V, is an AI video mode that takes two pictures: one the clip opens on and one it should end on. The model generates the motion between them from a prompt, so you decide where a shot starts and finishes and it invents the path.

Written and reviewed by the DreamBench team. Published 2026-10-29; updated 2026-10-29.

## First and last frame at a glance

- **Input:** Two pictures, a first and a last frame, and usually a prompt for what happens between them.
- **Output:** One shot, usually 4 to 15 seconds, opening on the first picture and steered to land on the last.
- **Fixed by you:** Both ends: the opening exactly, the ending as a target the model aims for.
- **The model decides:** The path: the motion, the camera, the timing, and anything neither picture shows.
- **Works best for:** Transitions, reveals, before and after, a product turning, a shot that must end on a logo.
- **Struggles with:** Two pictures that are too far apart, a change too big for the seconds, faces and text mid-move.

## How first and last frame video works

First and last frame works like [image to video](https://dreambench.ai/glossary/image-to-video) with a second anchor. The model places your first picture at the start of the clip and your last picture at the end, marks both as frames to keep, and generates the frames between them from noise until they join the two up in a way the prompt describes.

*Diagram: First and last frame: you fix both ends, and the model generates the frames between them.*

In [text to video](https://dreambench.ai/glossary/text-to-video) the model invents everything; with one picture, everything after the opening. With two it has a destination, so its job is finding a believable route: the mode suits a planned change and struggles when no such route exists.

#### Under the hood: two frames held, the middle generated

The report on the open [Wan](https://arxiv.org/html/2503.20314) model describes the usual design. The picture is compressed and placed as the first frame, with empty frames after it, and a binary mask marks which frames are kept and which are generated. An image encoder reads the picture’s content as well, so its look carries into frames it never showed. The report says the same design extends to first-last frame transformation and to video continuation.

The name FLF2V comes from that family: the Wan team released [Wan2.1 FLF2V](https://huggingface.co/Wan-AI/Wan2.1-FLF2V-14B-720P), first-last-frame-to-video, on April 17, 2025, at 720p only.

Research calls the task keyframe interpolation or inbetweening, after the animator’s job of drawing the frames between two key poses. One [ICLR 2025 paper](https://arxiv.org/abs/2408.15239) takes an image-to-video model trained to move forward in time from one picture, tunes a copy to move backward, and runs the two together so each step blends a guess from each end.

#### How to write a first-and-last-frame prompt

Describe the change, not the pictures: both are already fixed. Say what moves, in what order, and how the camera travels from the first framing to the last. Kling’s [guide for its O1 model](https://kling.ai/quickstart/klingai-video-o1-user-guide) gives the pattern: take the first image as the start frame, the second as the end frame, then describe the changes between them.

- Name one action that turns the first picture into the last: “she turns from the window to face the camera”.
- Name the camera move that gets from one framing to the other: “slow push in from wide to close”.
- Say how the clip arrives: “settles”, “comes to rest”, “the light finishes rising”, so the last second isn’t a snap.
- Leave out what both pictures already show: the face, the clothes, the room.

## First frame, first and last frame, reference or keyframes?

A picture can steer a video in several ways. Which you choose decides how much of the shot is yours and how much the model makes.

**Picture-led ways to make a shot, side by side**

|  | What you give | What is fixed | What the model invents | Pick it for |
|---|---|---|---|---|
| First frame only ([image to video](https://dreambench.ai/glossary/image-to-video)) | One picture | The opening, exactly | Everything after it, including where it ends | Bringing a photo or a designed frame to life |
| First and last frame | Two pictures | The opening, and the ending as a target | The path between them | Transitions, reveals, before and after, a fixed final image |
| [Reference to video](https://dreambench.ai/glossary/reference-to-video) | One or more pictures of a subject, a place or a product | Who or what appears, not where it stands | Every frame, including the first | The same character or product across many shots |
| Keyframes | Three or more pictures at points along the clip | Each of those moments | The paths between each pair | Choreographed motion with several beats; a few tools offer it |
| Video extension | An existing clip | Everything up to its last frame | What happens next | Making a shot longer than one generation |

First and last frame is the only one of these that fixes where a shot finishes; a longer sequence is several such shots, cut together.

## How to choose a first and a last frame that work together

Most failures start before the prompt, with two pictures that can’t be joined in the time. Kling’s [start and end frames guide](https://kling.ai/quickstart/ai-video-start-end-frames) asks for two similar images on the same theme, and warns that large differences may trigger a shot switch: a cut instead of a move.

- Give both pictures the same shape and size as the clip, so nothing is cropped or stretched at either end.
- Keep the subject, the light, the lens and the style the same; change one or two things, such as a pose, a camera position or the time of day.
- Make the change fit the length: a head turn fits in 5 seconds, a sunrise over a city does not.
- Make the last frame from the first: edit a copy with the [AI image editor](https://dreambench.ai/tools/ai-image-editor), or grow it to a wider framing with the [AI image expander](https://dreambench.ai/tools/ai-image-expander), so both share every detail you didn’t change.
- Keep faces, hands and text large and clear in both: the middle frames are invented, and small details are redrawn first.

## Start and end frames in Veo 3.1, Kling and Seedance

Each maker sets its own limits. What their pages said when we read them on October 29, 2026:

**First and last frame, in each maker’s own documentation**

|  | What the maker’s page says | Worth knowing |
|---|---|---|
| [Veo 3.1](https://dreambench.ai/models/veo-3.1) (Google) | [Google Cloud’s page](https://docs.cloud.google.com/vertex-ai/generative-ai/docs/video/generate-videos-from-first-and-last-frames) lists Veo 3.1 and Veo 3.1 Fast for first and last frames, with clips of 4, 6 or 8 seconds in 16:9 or 9:16 | Frames are JPEG or PNG; in Google’s Media Studio the end image is an optional slot under Image-to-video |
| [Kling 3.0](https://dreambench.ai/models/kling-3.0) (Kling AI) | [Kling’s 3.0 guide](https://kling.ai/quickstart/klingai-video-3-model-user-guide) lists Start & End Frames-to-Video for Kling 2.6 and 3.0, and 3.0 runs 3 to 15 seconds | In the Kling app it is Add End Frame on the Image to Video page; Kling’s O1 guide says an end frame alone is not supported |
| [Seedance 2.5](https://dreambench.ai/models/seedance-2.5) (ByteDance) | [BytePlus’s tutorial](https://docs.byteplus.com/en/docs/modelark/seedance-2-5): one image as the first frame, or two as the first and last frames; a first frame is required | The clip takes the first frame’s aspect ratio; each image 300 to 6,000 pixels a side and under 30 MB; no real human faces |
| Wan2.1 FLF2V (Wan-AI) | An open model released on April 17, 2025, 720p only | Trained mainly on Chinese text-video pairs, so its card recommends Chinese prompts |

## What people make with first and last frame

- Transitions: the same street by day and by night, a room empty and furnished, summer to snow.
- Before and after: a makeover, a renovation, a dish raw and cooked, both ends exactly as photographed or designed.
- Reveals that land on a planned final image: a product, a title card or a [logo](https://dreambench.ai/tools/logo-animation).
- A product turning from its front to its back, with both views exact.
- A character moving between two poses drawn for a storyboard.
- Loops, with the same picture at both ends.

## Which AI video models take an end frame

On DreamBench, [Seedance 2.5](https://dreambench.ai/models/seedance-2.5), [Seedance 2.0](https://dreambench.ai/models/seedance-2.0), [Veo 3.1](https://dreambench.ai/models/veo-3.1) with its Fast and Lite versions, [Kling 4.0](https://dreambench.ai/models/kling-4.0), [Kling 3.0](https://dreambench.ai/models/kling-3.0), [Wan 3.0](https://dreambench.ai/models/wan-3.0), [Wan 2.7](https://dreambench.ai/models/wan-2.7), [MiniMax H3](https://dreambench.ai/models/minimax-h3) and [LTX-2.3](https://dreambench.ai/models/ltx-2.3) can be told where to end. [HappyHorse 1.1](https://dreambench.ai/models/happyhorse-1.1), [Grok Imagine 1.5](https://dreambench.ai/models/grok-imagine-1.5) and [Vidu Q2](https://dreambench.ai/models/vidu-q2) start from a first frame only. The table is read from the same settings the DreamBench composer uses:

**Video models on DreamBench: sound, first and last frame, resolutions and aspect ratios**

| Model | Sound | First and last frame | Resolutions | Aspect ratios |
|---|---|---|---|---|
| Seedance 2.5 | On or off | Yes | 480p, 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 · 21:9 |
| Seedance 2.0 | On or off | Yes | 720p, 1080p, 4K | 1:1 · 16:9 · 9:16 · 4:3 · 21:9 |
| Veo 3.1 | Always on | Yes | 720p, 1080p, 4K | 16:9 · 9:16 |
| Kling 4.0 | Always on | Yes | 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 · 21:9 |
| Kling 3.0 | On or off | Yes | 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 · 21:9 |
| Wan 3.0 | Always on | Yes | 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 |
| Wan 2.7 | On or off | Yes | 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 |
| Grok Imagine 1.5 | Always on | First frame only | 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 |
| MiniMax H3 | Always on | Yes | 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 · 21:9 |
| HappyHorse 1.1 | Always on | First frame only | 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 · 21:9 |
| LTX-2.3 | On or off | Yes | 720p, 1080p, 4K | 16:9 · 9:16 |
| Vidu Q2 | On or off | First frame only | 720p, 1080p | 1:1 · 16:9 · 9:16 · 4:3 |

> “First frame only” means the model can open on your picture but can’t be told where to finish: the ending is left to the prompt.

## Why the end frame doesn’t match, and other limits

Why doesn’t my video end on my last frame? The first frame is placed exactly, but the last is a target: the model generates the final frames to meet it, and a path that runs late, or a change too big for the seconds, lands near it or snaps into it in the last moments. The closer the two pictures, the closer the landing.

**First and last frame troubleshooting**

|  | Likely cause | What to try |
|---|---|---|
| The clip cuts to the end frame instead of moving there | The pictures are too different, so the model joins them with a cut | Bring them closer in subject, light and framing, or make it two shots with a picture in between. |
| The last second snaps or morphs into place | The motion didn’t reach the end in time | Ask for a smaller change, a longer clip, or a motion that slows and settles before the end. |
| The final frame looks a little different from your picture | The last frame is generated to match, not copied | Run it again, or hold your own picture for a moment after the clip in an editor. |
| One end is cropped or stretched | The pictures and the clip differ in shape | Crop both pictures to the clip’s shape before you start, and give them the same size. |
| A face or a logo blurs in the middle | Only the two ends are fixed; every frame between is invented | Keep it large and similar in both pictures, move the camera less, and keep the clip short. |
| The motion takes an odd route | The prompt doesn’t name the path | Describe one action and one camera move that turn the first picture into the last. |
| The end frame is ignored | The model takes a first frame only | Pick a model that takes a last frame: the table above shows which. |

- Middle frames can still flicker or drift: a failure of [temporal consistency](https://dreambench.ai/glossary/temporal-consistency), worse the further apart the ends.
- Two ends fix one shot. A person who must look the same across several shots needs the same pictures in each, which is [character consistency](https://dreambench.ai/glossary/character-consistency).
- Some models refuse some pictures: ByteDance’s tutorial says Seedance 2.5 takes no images showing real human faces.
- Only use pictures you have the right to use, and ask before you animate a photo of someone else.

## First and last frame on DreamBench

The [logo animation](https://dreambench.ai/tools/logo-animation) tool is DreamBench’s first-and-last-frame run. You upload a logo and pick a style, and the model is given your logo, as you uploaded it, on its own backdrop, as the frame to end on. The first frame depends on the style: the logo itself for Light sweep, the empty backdrop for a reveal, or a frame drawn from the logo, broken up for Glitch, a dim outline for Neon, edge-on for 3D spin.

Each run is a five-second clip on [Kling 3.0](https://dreambench.ai/models/kling-3.0): 720p and silent, 1080p, or 1080p with sound. Some runs don’t land cleanly on the logo frame; if the last second looks wrong, run it again.

The [image-to-video tool](https://dreambench.ai/tools/image-to-video) is the one-picture version: a photo becomes the first frame of a five-second clip. And when you [describe a film](https://dreambench.ai/), DreamBench writes it as shots, draws an opening frame for each from your reference pictures, and the video model generates the shot from it; where it goes is written in its prompt.

## Questions

- **What does FLF2V mean?** FLF2V is short for first-last-frame-to-video: generating a clip from two pictures, one it opens on and one it ends on. The name was used by the open Wan2.1 FLF2V model, released in April 2025; most apps call the same mode start and end frame.
- **How do I use first and last frame in Kling?** In the Kling app, open Image to Video, click Add End Frame, upload your start image and your end image, describe the change between them and generate. Kling’s guide lists start and end frames for Kling 2.6 and 3.0, asks for two similar images on the same theme, and warns that large differences may trigger a shot switch.
- **Can Veo 3.1 do start and end frame?** Yes. Google Cloud’s documentation lists Veo 3.1 and Veo 3.1 Fast for generating a video from a first and a last frame, with clips of 4, 6 or 8 seconds in 16:9 or 9:16. On DreamBench, Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite are all marked as taking a last frame; Veo 3.1 is available on Plus and above.
- **Does Seedance 2.5 support first and last frame?** Yes. ByteDance’s tutorial for Seedance 2.5 takes one image as the first frame, or two images as the first and last frames; a first frame is always required, and the clip takes the first frame’s aspect ratio. It does not take images showing real human faces.
- **Why does my end frame not match?** Your end frame doesn’t match because the model treats it as a target, not a frame to paste in: it generates the last frames to meet it, and if the change is too big for the seconds, or the motion runs late, it lands near the picture or snaps into it at the end. Make the two pictures closer, ask for a smaller change or a longer clip, and describe a motion that settles before the end.
- **Why does my first and last frame video cut instead of moving?** A first and last frame video cuts instead of moving when the two pictures are too different to join with believable motion, so the model joins them with a shot change. Kling’s own guide warns that large differences may trigger a shot switch. Keep the subject, light and framing similar, change one or two things, or split the change into two shots.
- **Can I give only an end frame?** Usually not. The models documented here need a first frame, and the last is the optional one: Kling’s O1 guide says generation with only an end frame is not supported, and Seedance 2.5 requires a first frame. To make a shot that ends on a picture, give a first frame as well, even a plain background for a reveal.
- **What is the difference between first and last frame and keyframes?** First and last frame fixes the two ends of one shot. Keyframes fix three or more moments along a clip, and the model fills each gap between them. Researchers also call first and last frame keyframe interpolation, since it is the two-keyframe case.
- **Is first and last frame the same as frame interpolation?** No. Frame interpolation adds frames between two frames that are a split second apart, to make motion smoother or slower, and invents almost nothing. First and last frame joins two pictures that can be seconds apart in time, so the model generates the whole motion between them from a prompt.
- **Can I make a looping video with first and last frame?** Yes, on a model that takes a last frame: give the same picture as both ends and describe a motion that goes out and comes back. A small, slow motion loops best; the join can still jump slightly if the clip lands near the picture rather than on it.
- **Do the first and last frame need the same aspect ratio?** They should have the same shape and size as each other and as the clip. A picture of another shape is cropped or stretched to fit, which can move the subject between the two ends; Seedance 2.5, for example, gives the clip the first frame’s aspect ratio. Crop both pictures to the clip’s shape before you start.
- **Which AI video models support start and end frames?** On DreamBench, Seedance 2.5, Seedance 2.0, Veo 3.1 with its Fast and Lite versions, Kling 4.0, Kling 3.0, Wan 3.0, Wan 2.7, MiniMax H3 and LTX-2.3 take a last frame. HappyHorse 1.1, Grok Imagine 1.5 and Vidu Q2 start from a first frame only. The table on this page is read from the composer’s own settings.
- **How long can a first and last frame video be?** One shot, as long as the model makes: Google lists 4, 6 or 8 seconds for Veo 3.1, and Kling 3.0 runs 3 to 15 seconds. DreamBench’s logo animation, which ends on your logo, is always five seconds. For longer, plan several shots, each with its own two ends, and cut them together.

## Sources

- [Google Cloud: Generate videos using first and last video frames](https://docs.cloud.google.com/vertex-ai/generative-ai/docs/video/generate-videos-from-first-and-last-frames), Google Cloud: read 2026-10-29
- [Kling AI: Video Start & End Frames Guide](https://kling.ai/quickstart/ai-video-start-end-frames), Kling AI: read 2026-10-29
- [Kling AI: Kling VIDEO 3.0 Model User Guide](https://kling.ai/quickstart/klingai-video-3-model-user-guide), Kling AI: read 2026-10-29
- [Kling AI: Kling VIDEO O1 User Guide](https://kling.ai/quickstart/klingai-video-o1-user-guide), Kling AI: read 2026-10-29
- [BytePlus ModelArk: Dreamina Seedance 2.5 tutorial](https://docs.byteplus.com/en/docs/modelark/seedance-2-5), BytePlus: read 2026-10-29
- [Wan: Open and Advanced Large-Scale Video Generative Models, §5.1 Image-to-Video Generation](https://arxiv.org/html/2503.20314), arXiv: read 2026-10-29
- [Wan2.1 FLF2V 14B 720P model card](https://huggingface.co/Wan-AI/Wan2.1-FLF2V-14B-720P), Wan-AI: read 2026-10-29
- [Generative Inbetweening: Adapting Image-to-Video Models for Keyframe Interpolation](https://arxiv.org/abs/2408.15239), arXiv: read 2026-10-29

DreamBench is an independent platform and isn’t affiliated with or endorsed by the companies named on this page.