How AI image to video works.

Firefly's Image to Video gives you directorial control without needing a cinematographer. Your first frame is the image you start from. The optional last frame is where you want the scene to end. The prompt describes the motion in between.

Adobe's own video model does the rest. It reads the image, determines how the elements in it could move, and generates a short video clip up to a five-second clip at 1080p resolution.

The output is a real video clip rather than an animated still. Subjects can move, cameras can pan or zoom, and the atmosphere can shift across the duration of the clip. Output is MP4, sized to your chosen aspect ratio, and ready to use in a longer edit or drop into a Reel, Short or promo.

This guide walks you through creating a clip from images from start to end. It explains how the workflow runs, what to put in your prompt and what to do when the first generation isn't quite right.

The Firefly images to video workflow.

Watch the video for a quick view of what's possible and follow up with the detailed steps when you're ready to start.

  1. Open the Image to Video workspace. Sign in at firefly.adobe.com. Select Video under All features and then Image to video.
  2. Choose your AI model. Firefly Video Model is the default. Partner video models are also available inside the same workspace, with their own strengths and pricing. See Firefly partner models for the current line-up and the commercial-use details that apply to each.
  3. Set your resolution and aspect ratio. Pick the format you'll publish in: 9:16 for Reels, Shorts and TikTok; 16:9 for YouTube; 1:1 if you need a square crop. Decide here rather than later because some detail can be lost in cropping.
  4. Upload your first frame. This is the image your video starts from. A clear subject, sharp focus, good lighting, and 1080p or higher all help – the better the source, the smoother the motion prediction. If you don't have a source image ready, you can generate one with Text to Image and bring it straight in.
  5. Upload the final frame (optional). Upload a second image to tell Firefly where the scene should end. When a final frame is used, Firefly interpolates between the two frames. Skip this step and Firefly works from the first frame and your prompt alone.
  6. Write a motion prompt. Describe the motion you want to see, covering the camera movement, subject motion, atmosphere and visual style.
  7. Generate the clip. Click Generate and Firefly quickly returns a fully animated clip. Download to save the MP4 to your computer and you're done.

How to write a motion prompt that works well.

A good motion prompt for Image to Video has four parts: subject motion, camera movement, atmosphere and visual style.

Subject motion

Describes what the things in your image do. Be specific and concrete. Use action words to clearly describe the movement.

  • A mouse jumps over moss, raising dust.
  • A model turns her head towards the light.
  • A paper boat drifts forward in still water.

Camera movement

Tells Firefly how the camera should behave. Use camera terms for precision – "dolly left" rather than "move slowly to the left to reveal the scene". Including even one camera term materially changes the result.

  • Slow push-in.
  • Dolly left.
  • Tilt up to reveal the sky.
  • Static handheld.

Atmosphere

Sets the mood and lighting of the clip. Atmosphere can be the difference between a clip that's impressive and one that's just OK.

  • Golden hour.
  • Overcast and still.
  • Misty and cinematic.

Visual style

Anchors the look – but it's important that the style matches the source image.

  • Photorealistic.
  • Cinematic.
  • Anime.
  • Watercolour.
  • Claymation.

Remember to keep a consistent style. A single prompt can't make a reliable claymation clip from a lifelike photo.

Here's a worked example, building on the four parts.

Flower bud in the foreground opens. Slow push-in. Soft light, cinematic.

That's three precise sentences (a dozen words) and it gives Firefly enough direction to produce a coherent five second clip without ambiguity.

If the first generation isn't quite right, add more detail or change the prompt and try again. You can also use Prompt to Edit to change one element without regenerating the whole thing. This is a useful way to remove a distracting object, swap the background or adjust the motion without losing the parts of the clip that worked.

Tips for better results.

A handful of patterns separate the AI generations that work from the ones that don't.

  • A high-resolution source image. Anything blurry, low-light or compressed gets compounded by the motion prediction.
  • A clear subject. Busy backgrounds and low-contrast subjects confuse the AI. A clean subject with a recognisable silhouette produces noticeably smoother motion.
  • A precise prompt. Overloaded prompts produce muddier output. Stick to one main piece of subject motion, one camera move and one atmospheric note. If you want two things to happen, run two generations rather than asking for both in one.
  • A matched style. If your image is photorealistic, prompt for photorealistic motion. If your image is illustrated or anime-style, prompt for the same. Crossed signals produce flicker.
  • Small iterations. Change one element of the prompt between generations rather than rewriting the whole thing. This will help you isolate elements that may be problematic.

If you're consistently getting results that aren't quite right, change your source image and try again to see if the same prompt produces a better result.

What to do next with your clip.

A five-second clip is rarely the final deliverable. Firefly's video toolkit handles the next steps in the same workflow.

  • Add a soundtrack. Image to Video output is silent, so bring it to life with Generate Soundtrack to create a custom track that fits the clip's pace and mood.
  • Assemble a longer story. Firefly Video Editor is a browser-based multi-track timeline space, currently in public beta, built for blending generated clips with your own footage. You can also send a clip straight into Premiere Pro or Adobe Express for further work.
  • Translate it for another audience. If your clip needs to reach viewers in another language, Translate Video handles that without losing the original voice.

Using Firefly and the included Adobe Express features, you can build on one image-to-video clip in different ways. It's easy to make a Reel, a YouTube ad cut and a longer brand video, each with a slightly different downstream treatment.

Make your first video from an image.

The way forward is easy. A clear source image. A focused motion prompt. The right aspect ratio for where it's going to live. Firefly does the rest in a few seconds, and the output is yours to use on your socials, in personal projects, or commercially with Firefly Video Model.

Frequently asked questions about making a video from an image.

Share this page

You may also like