The short version: you upload a photo into a tool like Runway, Kling, Luma Dream Machine, or Google’s Veo, write a short prompt describing the motion you want, pick a duration and camera movement, then generate and pick your best take from several attempts. It takes minutes, it costs a few pounds per clip, and it rarely works perfectly on the first try, whatever the demo videos tell you.
Why I even started doing this
Last spring a client sent me one product photo for a kitchen gadget, a single flat shot on a white background, and asked for “something for Facebook that moves a bit, we don’t have budget for a proper video shoot.” That’s how most of my clients arrive at image-to-video AI. Not because it’s exciting new tech, but because they’ve got one decent photo and no budget for a camera crew.
I ran that photo through Runway’s Gen-3 model, wrote a prompt asking for a slow rotation and steam rising from the gadget’s spout, generated four versions at 5 seconds each, picked the one where the steam didn’t look like smoke from a house fire, and had a usable ad clip inside twenty minutes. That’s roughly what this process looks like every time now, image in, several attempts, one keeper.
The tools that do this well
There are four I reach for depending on the job:
- Runway (Gen-3 and Gen-4) – the most controllable, with camera direction options like pan, zoom, and tilt built into the interface. Good for product shots and anything where you need the motion to look deliberate.
- Kling AI – strong at human motion, so if your still image is a person and you want realistic movement (hair, clothing, subtle expression change), Kling handles it better than most.
- Luma Dream Machine – fast and free to start, decent for atmospheric shots like landscapes, food, or interiors where you just want gentle drift rather than complex action.
- Google’s Veo (inside Gemini and Flow) – the newest of the big four, and the one improving fastest on physics, meaning objects fall and bounce more like they should rather than melting.
I’ve tested all four on the same brief more than once and the honest pattern is this: no single tool wins every category. Runway for product, Kling for people, Luma for mood, Veo for anything with physical action like liquid pouring or fabric moving. There’s a closer Kling vs Runway comparison worth a look if you want to dig into just those two.
The actual step-by-step process
Here’s what I do, tool-agnostic, because the steps are nearly identical across all of them:
- Step 1: Start with the cleanest image you have. High resolution, good lighting, no watermark, no busy background. Garbage in, garbage out applies harder here than in almost any other AI task.
- Step 2: Write a motion prompt, not a scene description. The AI already has the scene, it’s your still image. Your job is to describe what should move and how. “Camera slowly zooms in, steam rises gently from the cup, soft daylight” works far better than “a cozy coffee shot.”
- Step 3: Set your duration. Most tools default to 5 or 10 seconds. For social ads I almost always use 5, because longer clips give the model more time to drift into strange territory, extra fingers, warped edges, objects that flicker.
- Step 4: Choose camera movement if the tool offers it. Pan, zoom, orbit, static with subject motion only. Pick one, don’t stack three movements or you’ll get a seasick result.
- Step 5: Generate multiple versions. I budget for at least four attempts per image. This isn’t a failure of the tool, it’s just how the generation works, the same prompt on the same image gives different results each run.
- Step 6: Review at real size, not thumbnail. Faults hide in small previews. Play it full screen before you approve anything for client use.
- Step 7: Export and, if needed, extend. Some tools let you take the last frame of your generated clip and run it again as a new starting image, effectively chaining clips into a longer sequence.
What it costs
This bit gets glossed over a lot, so here are real numbers. Runway’s paid plans run from roughly £12 a month for a basic tier up to £60-plus for unlimited generations on their standard model, with the higher-quality Gen-4 pulling credits faster, often 2 to 5 credits per second of video depending on resolution. Luma Dream Machine has a usable free tier, then paid plans starting around £8 to £10 a month. Kling operates on a credit system too, with free daily credits that run out fast if you’re testing prompts repeatedly. If you’re doing this for one campaign, budget for maybe £20 to £30 and a few hours of trial and error rather than expecting a single free generation to solve the brief.
The uncomfortable bit nobody wants to put in these guides
Here’s what most posts on this topic won’t tell you plainly: these tools are impressive and also unreliable, at the same time, in the same session. You will get a beautiful clip on attempt two and then three broken, uncanny ones on attempts three, four, and five with the exact same prompt. Hands still warp. Text in the background of your image often smears into nonsense. Reflections in glass or water frequently do something physically impossible. I’ve had a coffee cup grow a second handle mid-clip. This isn’t a knock against the technology, it’s moving fast and getting better every few months, but if you’re planning client work or a campaign timeline around “generate one clip, done,” you’re setting yourself up to miss a deadline. Build in the review and retry time. Treat it like a slot machine that pays out reasonably often, not a vending machine that gives you exactly what you asked for every time.
Where these clips work best
Not every still-to-video output belongs on every platform, and knowing where to put the finished clip matters as much as making it. Short 5 to 10 second clips with subtle motion do well as scroll-stopping Facebook content that turns followers into buyers, especially when the motion is small enough that it doesn’t look obviously artificial at a glance. If you’re running paid traffic, that same kind of subtle-motion clip has lifted click-through for a couple of my clients when swapped in for a static image, because it interrupts the scroll pattern, something I’d factor into anything you’re building around a Facebook ad campaign designed to convert leads into customers. Pinterest is another underrated home for these, since Idea Pins reward motion and a still-turned-video product shot performs noticeably better there than the flat original, which is one more reason I still think Pinterest is worth it for small business in 2026.
Turning one image into a week of content
The real value isn’t the single clip, it’s what you can build around it. I’ve taken one hero product photo, generated three or four different motion variations of it (a slow zoom, a rotation, a close-up detail pan), and used those across a full week of posts, an Instagram Reel, a Pinterest Idea Pin, a Facebook ad, and a LinkedIn post. It’s a similar principle to how you’d stretch a single podcast recording into multiple assets, and if you haven’t done that exercise yet it’s worth reading how to repurpose a podcast episode into a week of content, the mindset transfers directly to images. If you want the fuller toolkit for pulling this off across formats, I keep an updated list of the best content repurposing tools for video, audio, written content and social in 2026, and a separate rundown of five AI video tools for making high-converting videos if you want to compare options beyond the four I’ve named here.
Prompts that work (and ones that don’t)
Vague prompts produce vague, often broken motion. Here’s the difference in practice:
- Weak: “make this photo come alive”
- Better: “gentle handheld camera drift left to right, hair moves slightly in a light breeze, subject blinks once”
- Weak: “add some action”
- Better: “liquid pours from the top of frame into the cup, camera stays static, steam rises after two seconds”
Specificity about what moves, in what direction, and roughly when, gets you closer to a usable clip on fewer attempts. I keep a running note of prompts that worked well for specific image types (products, faces, landscapes, food) because the phrasing that gets good steam doesn’t always get good hair movement.
A quick sanity check before you use any clip publicly
Play the full clip at least twice, once at normal speed and once slowed down if your player allows it. Check hands, text, reflections, and edges of the frame first, that’s where the model usually slips up. If it’s for a client, get a second person to watch it cold, without you pointing out what to look for, because you’ll have gone blind to the small wrongness after staring at the same file for twenty minutes.
Frequently asked questions
Can you turn any photo into a video with AI?
Mostly yes, but clean, well-lit, high resolution images with a clear subject give far better and more predictable results than dark, cluttered, or low-resolution ones, and photos with visible text or complex reflections tend to produce more errors than simple product or portrait shots.
How long does it take to turn a still image into an AI video?
The generation itself usually takes one to three minutes per clip, but expect to run four or more attempts before you get a clean result, so realistically budget twenty to forty minutes per image including review time.
Is AI image-to-video good enough for real client work?
Yes for short 5 to 10 second social clips, product ads, and background motion, but it’s not yet reliable enough for anything requiring precise, repeatable human motion or long-form footage without visible flaws, so match your expectations to short-form use cases first.
What’s the cheapest way to try this before paying for a tool?
Luma Dream Machine and Kling both offer free daily credits that let you test the process on a real image before committing to a paid plan, which is the easiest way to judge whether the output quality matches what your project needs.
Official documentation
Related reading: How to Write Better AI Image Prompts (What I Learned After 200+ Bad Ones) and How AI Image Generators Turn Your Text Prompts Into Pictures.
This builds on my main AI marketing guide, my main guide on the topic.
If you want your own guide on this published here, you can write for us about productivity.
