You decide exactly where the clip starts, and optionally where it ends. It's the go‑to mode for animating product shots and building before/after transitions.
When to use this mode
You have a product photo and want to bring it to life
You want a smooth transition between two images
You need precise control over how the video opens and closes
Steps
Go to All agents → Image to Video or go to Generate
Ensure you've selected the video tab
Upload your starting image using the uploader
This becomes your first frame
(Optional) Add an end image: This becomes your last frame
Write your prompt describing how the image should animate
Set your aspect ratio, duration, and model
Click Generate
Your video appears in the same window in about 5-8 minutes. You can queue a second variation while the first renders.
First frame, last frame, reference image, what's the difference?
This is the concept that makes or breaks image‑to‑video. These three are not interchangeable.
Think of directing a driver:
A first frame is "start at this exact intersection."
First + last frame is "start here, end there, and pick the route between."
A reference image is "this is the kind of place I like."
First frame
The literal opening shot. It locks composition, lighting, and subject. The quality of your first frame sets the ceiling for the whole video; always start with a clean, high‑resolution image. Use it when the clip should begin exactly on your image.
Last frame
The target ending the model interpolates toward. It's a strong directional guide, not a pixel‑perfect lock. Use it when you want a transformation, a product before/after, a branded outro, or a loop. This is what makes animating between two images possible. Two rules for first + last frame:
Keep the same scene and subject between the two images, or you'll get morphing artefacts
Match the lighting and color between them, or the clip will shift color mid‑way
Reference image
Guides identity, style, or content without being tied to any frame. The model decides how to weave it in. Useful for holding a product's look or a character's identity across shots. Reference images are available on specific models only. If you switch models and the option disappears, that's expected — the rest of this guide still applies.
The general rule: you choose either frame control or reference images in one generation — not both.
Writing the prompt
The image already shows the subject — so don't re‑describe it. Spend your prompt on movement and camera instead. Image‑to‑video prompts are shorter than text‑to‑video prompts.
Template
[One or two simple movements]. [One camera movement]. Preserve composition and colors. Avoid [things to exclude].
Examples
Animating a product shot:
The perfume bottle rotates slowly on its base, light glinting off the glass. Camera holds a steady close-up. Preserve composition and colors. Avoid warping the label.
A two-image transition (first frame: empty scene · last frame: styled scene):
Smoothly reveal the styled scene from the empty one. Gentle push-in. Soft, consistent lighting throughout. Avoid abrupt jumps.
For product fidelity, always add "preserve composition and colors." It tells the model to keep your product looking like your product.
Get the best result
Keep movements simple and focused: one or two actions work best.
Use high‑quality source images for smoother results.
Start with shorter durations to test your prompt before committing to longer, costlier ones.
Don't re‑describe the subject: the image already defines it. Describe the motion.
Avoid complex multi‑step animations in a single prompt.
Want a consistent character that moves and speaks?
If your "image" is a person you want to present and talk across multiple videos, use an avatar instead of a plain image: it keeps the same face and voice every time. See Avatar to Video.
Common mistakes
A low‑quality first frame. Everything downstream inherits its flaws
Mismatched first and last frames. Different subjects morph; different lighting shifts color. Keep them consistent
Re‑describing the product in the prompt. The image handles that — describe the movement
Too many actions. One or two clean movements beat a busy, multi‑step animation
Credits
The cost is shown next to the Generate button before you commit, and scales with model × duration × resolution. Adding a second image or other references can change the cost. Failed generations are refunded automatically. Start with one short output to test your prompt.