Turn a photo into a video, three ways

Start with a still worth animating

A video model animates what it is given. A sharp photo with one clear subject, light that comes from somewhere, and room around the subject to move gives every path below a head start. A soft, cluttered or heavily filtered photo makes the model guess, and guessing is where limbs bend and faces smear.

We made this one on the image composer with Nano Banana Pro at 16:9. The prompt names the wardrobe, the place, the two light colours and the lens, and forbids text and logos, so nothing in the frame is a brand.

Fashion film still, 16:9, photoreal. A young East Asian woman with short bleached platinum hair stands alone on a flooded rooftop car park at night in heavy rain, wearing a silver metallic puffer jacket zipped up to the chin, black leather gloves and black trousers, hands relaxed at her sides. A cyan gel light from the left and a magenta gel light from the right, thick haze, a shallow sheet of rainwater at her feet reflecting both colours, rain streaks lit from behind. She looks just past the camera, calm and certain. Medium shot, 50mm lens, slightly low angle, shallow depth of field, hard contrast, deep blacks, film grain. An invented person, not a real person. No logos, no brand marks, no text, no signage.

What goes wrong

Before you start

• Have the photo as a PNG, JPEG or WebP. The composer's upload takes exactly those.
• Decide the aspect ratio from the photo, not from the platform you will post on. Reframe afterwards in Studio if you must.
• Know the one move you want: a turn, a step, a breath, a camera push. Write that sentence before you open anything.
• Read the price on the Generate button before you press it. It is the exact charge for the settings you picked.
• Every path here works on Pro and up: the models that animate a photo sit on those plans.

Path 1, most control: the video composer

Open the video composer and choose Animate an image. Drop the photo in, or pick it from your library. Choose the model, set the aspect ratio, the duration and the quality, write the motion prompt and press Generate. The button shows the price for exactly those settings.

We used Seedance 2.5 at 720p, five seconds, 16:9, with audio on. It takes the photo as the first frame, so the woman, the rain and the lights are yours, and the model only has to move them.

Animate this still. The rain keeps falling and the haze drifts slowly through the cyan and magenta light. She stands still, breathes once, then turns her head to look straight into the lens. The water at her feet ripples with the rain and mirrors the lights. Slow push in from a low angle, the camera rising a little as it moves. No cuts, no new people, no text. Sound: heavy rain on wet concrete, a distant city hum.

Path 2, hands off: ask the agent

Open the agent, attach the photo with the paperclip and type one line. There is no plan to approve: the agent picks a model that takes a photo, writes the motion prompt itself, runs it, and answers with the finished clip and a link to it in your library. We asked for a draft tier and it took us at our word.

Animate this photo. Keep the rain, the haze and the coloured lights. She turns to look into the lens, slow push in, about five seconds, with sound. A draft quality tier is fine.

It chose H3 Max, the image variant, at the lowest quality tier, 5 seconds, and wrote this motion prompt from our line and the photo:

A slow, smooth push in on the woman standing in the falling rain under teal and magenta neon lighting on the wet rooftop. She turns her head to look directly into the camera lens. Rain and haze fill the atmosphere. Ambient sound of falling rain and city night ambience.

Path 3, fastest: a preset

Cover reveal is the preset that takes exactly one still and animates into it: the scene comes alive first and the clip lands on your photo as its last frame. Give it the still, pick how it moves (Slow push in, Parallax drift, Elements assemble, Light sweep, Focus pull or Living scene), pick the length, sound or silent, and whether it lands on the still or loops.

Make yours opens a guided chat that asks for one thing at a time. We gave it the still, skipped the words, chose Living scene and left the length, sound and ending at their defaults, which are five seconds, with sound, landing on the still. The review card shows every answer and an estimate for the clip before you press Create. Then the agent behind the preset writes the full prompt, renders the start frame and the clip, and replies with the result.

Which models take a photo

Every video model on NOLGIA publishes what it accepts, and the composer reads that list live. Three things matter for a photo: a start frame (your photo is frame one), an end frame as well (the clip must arrive at a second photo), and reference images (the model looks at your photo and composes its own shot, so the first frame is not your picture). This is the catalog on the day of writing; the models page is the live version.

Write the motion, not the picture

• Start with the verb. Turns, steps, breathes, looks up. The still is the noun.
• One camera move. Say it in the words the camera chip row uses: push in, orbit, crane up, tracking, or locked off. The composer's camera move chips append the exact fragment at the strength you pick, so you can leave the move out of your text and pick a chip instead.
• Say what stays. No cuts, no new people, no text. The model treats a missing rule as permission.
• Give the environment a slow life. Rain keeps falling, haze drifts, water ripples. It sells the clip without competing with the subject.
• Put sound last. One line that names the room and the loudest thing in it.

What it costs

You pay per clip, by length and quality tier, and the exact price sits on the Generate button before you press it. A request refused before the model runs costs nothing. The agent and the preset also bill the agent's turn, and the preset adds an image render for its start frame; its page quotes the whole run. These are the models the three paths used on our run: Seedance 2.5 in the composer and inside the preset, H3 Max from the agent.

Where you still regenerate

• The turn. Three paths, three readings of the same sentence. The composer clip turns her eyes to the lens late; the agent's clip pushes in close and lands the look; the preset holds her pose and lets the scene breathe, because its job is to return to the still. If the turn is the point, write it as the first thing that happens, with a time marker.
• Hands and gloves. Fingers are the first thing to go wrong when a hand crosses the body. Keep hands still or out of frame on the first run.
• The reveal's lettering. Cover reveal is built for stills that carry words. Leave the words field empty for a photo with none, and check the last second for invented type.
• The agent's model choice. It picks a model that takes a photo and tells you which. If you wanted a specific one, say so in your line.

Checklist

1. One sharp photo, PNG, JPEG or WebP, with room around the subject.
2. Aspect ratio of the clip matches the photo.
3. A model from the table that takes a start frame.
4. A motion prompt that names one action, one camera move, what stays, and the sound.
5. The price on the button read before you press it.
6. Play the clip full screen once before you post it, and watch the hands.