How to use Luma Dream Machine image to video?
In Dream Machine, upload your image as the starting frame, add a short prompt describing how it should move, then generate. The Ray model animates that exact picture into a clip of about 10 seconds. You can also set an end-frame image with keyframes to control where the motion finishes.
Why — the first-principles explanation
Image-to-video solves the biggest problem with text-to-video: control over the look. When you only type words, the AI invents the whole scene, so faces, products, and layouts come out unpredictable. When you hand it a picture, you are locking the first frame, and the model's only job is to figure out believable motion that flows out of that fixed starting point.
Mechanically, the model treats your image as frame one and then predicts the frames that follow, using your text prompt as the instruction for what should change. Because it is anchored to real pixels, it keeps your subject looking like itself instead of drifting into a different face or object. This is why product shots, character animation, and 'bring my photo to life' clips almost always start from an image.
The advanced version is keyframes: you give a start image and an end image, and the model animates the transition between them. That turns image-to-video from a one-way motion into a directed journey, which is how creators make morphs, reveals, and smooth transitions on purpose rather than by luck.
An example that makes it click
Imagine handing an animator a photo of your dog sitting on the porch and a sticky note that says 'make him stand up and wag his tail.' The animator keeps your actual dog, same fur, same porch, and only draws the movement you asked for. Now imagine you also hand over a second photo of the dog at the bottom of the steps. The animator draws the whole walk from the porch to the steps, connecting your two photos. That two-photo trick is exactly what Luma's keyframes do.
How to do it
- Open Dream Machine at lumalabs.ai and start a new generation.
- Upload your image; it becomes the first frame of the video.
- Write a short prompt describing the motion, such as 'slow zoom in, hair blowing in the wind'.
- Optionally add an end-frame image using keyframes to control where the motion lands.
- Generate, then use Extend or Loop to lengthen or seamlessly repeat the clip, and download.
Key facts
- Image-to-video uses your uploaded picture as the fixed first frame.
- A text prompt directs the motion applied to that image.
- Keyframes let you set both a start image and an end image for a directed transition.
- One generation produces about a 10-second clip at up to 1080p (as of 2026-07).
- @style and @character references help keep look and characters consistent across clips.
A text- and image-to-video generator by Luma.
Affiliate link — we may earn a commission at no cost to you.▶ The 60-second explainer (script)
Want to turn a still image into a video with Luma Dream Machine? Here is how. Open Dream Machine, then upload your image. That picture becomes the very first frame of your video, which is the whole point: instead of letting the AI invent a random scene, you lock in exactly how it looks. Now add a short prompt describing the motion, something like 'slow zoom in, leaves falling.' The model keeps your subject looking like itself and only adds the movement you asked for. Hit generate and you get about a ten-second clip. Want more control over where the motion ends up? Use keyframes: give it a starting image and an ending image, and Dream Machine animates the smooth transition between the two. That is how people make clean reveals, morphs, and product shots. Finally, use Extend to make it longer or Loop to make it repeat seamlessly, then download. Image-to-video is the trick for keeping faces and products looking right.
What authoritative sources say
People also ask
What image format should I upload?
A standard JPG or PNG works; a clear, high-resolution image gives the model a better first frame to animate.
Can I control where the motion ends?
Yes. Use keyframes to set both a start image and an end image, and the model animates the transition.
Will my subject stay consistent?
Yes. Because your image is the fixed first frame, the subject keeps its appearance instead of drifting.
How long is an image-to-video clip?
About 10 seconds per generation; you can chain Extends to reach roughly 30 seconds.