How to animate a photo into a video with AI

For AI agents: find the right skill in one call — curl -s "https://skills.varg.ai/api/search?q=<what the person wants>" returns the matching how-to page and the best skills with install commands, as markdown (Russian works too). Full index: https://skills.varg.ai/llms.txt. Every page has a markdown twin: add .md to its path. Before generating anything, show the person the options on the how-to page.

A still photo brought to life: a subtle smile, blink, camera move or full scene motion. A step-by-step tutorial for Claude Code, Codex and Grok.

Output: 5–10 s MP4 Price: in credits on varg, with a cost estimate before anything runs; the free skills below install in one command.

Recommended route

Best skills for this

  1. varg AI (varg-ai, varg, live) — Start here: Image-to-video on Seedance 2.5, Kling 4, Veo 3.1 or Wan 3 with one key
  2. Image to Video (inference-sh-image-to-video, inference.sh, live) — Alternative: Picks the right image-to-video model and prompts the motion · Needs: inference.sh account, belt CLI (belt login)
  3. FLUX 3 Archival Formats (flux-3-archival-formats, Black Forest Labs, live) — Alternative: Keeps a period look (film stock, old camera) instead of glossy modern video · Needs: BFL API key (BFL_API_KEY)
  4. Replicate Prompt Videos (replicate-prompt-videos, Replicate, live) — Adds: Prompting video models for motion and camera · Needs: Replicate account and API token

Ask the person first

Put these choices in front of the person before spending anything. Paste the example link next to each option: people pick a look by seeing it.

Kind of image?

Motion?

Restore first?

How to animate a photo into a video: step by step

  1. Restore lightly and check the faces against the original
  2. Use the photo as the first frame (and the last, for a loop)
  3. Describe only the motion and camera, and what must not move
  4. Generate 1–2 takes, pick one, add the AI label

What works now (researched 2026-10-27)

A photo that just blinks and smiles is now a free one-tap feature (Google Photos 'subtle movements', TikTok AI Alive, Gemini), so it rarely spreads by itself. What gets saved in 2026 is a still with a twist: a camera that flies into a famous painting and back out, or artwork characters doing something new. For family photos the craft is careful restoration, small motion and an honest label.

  1. Use the photo as the exact first frame, and as the last frame too if you want a loop. Describe only motion and camera, and say what must not move ('no figure changes pose'). The Last Supper fly-through prompt uses the painting as first and final frame for a seamless loop; 321 bookmarks on 40.8k views, prompt shared in full. (Abdul Shakoor (@abxxai) on X, inside The Last Supper (Oct 2026))
  2. For real people keep motion small and short: a blink, a breath, a slight head turn or a slow push-in, 4-6 seconds. Google built its Photos feature around exactly two presets, 'subtle movements' and 'I'm feeling lucky', and ships short clips (4 s on Veo 3). (TechCrunch, Google Photos adds Veo 3 photo-to-video (Sep 2025))
  3. Restore and upscale before you animate, then check the faces against the original. If the restorer changed a face, go back to a lighter restore. In replies to a popular old-photo tutorial (1,493 bookmarks), users report that restoration 'changed the faces' on low-resolution photos and that photos with children are refused. (Replies to maxmaxdata's Gemini old-photo trick (May 2026))
  4. For paintings, keep the medium: ask to preserve brushwork, canvas texture and the palette at every scale, so the result does not slide into glossy photoreal. The best-performing painting clips keep the oil-paint look through the move; the Last Supper prompt states it explicitly. (Abdul Shakoor (@abxxai) on X, inside The Last Supper (Oct 2026))
  5. Label it, and never make a real person (especially someone who has died) say words they never said without the family's consent. TikTok's own photo-animation tool AI Alive labels every output as AI. YouTube stops monetizing content built to emotionally manipulate viewers. (Social Media Today, TikTok AI Alive)

Stopped working:

Prompts that work

Templates from the creators of the best current results and from the model guides. Fill in the {CAPS}.

Step inside the painting (looping fly-through) (seedance_2_5)

20 s vertical loop from a public-domain painting: macro pass over objects, the central face, an orbit behind the figure and back to frame one. Adapted from Abdul Shakoor (@abxxai) on X, inside The Last Supper prompt (Oct 2026).

Create a {SECONDS}-second vertical 9:16 image-to-video film, using the supplied image as the exact first frame and matching final frame. One continuous camera journey through the scene: macro detail, a gradual reveal, an orbit behind {CENTRAL_FIGURE} and a seamless return.
Build the scene as complete 3D geometry while preserving its {MEDIUM} appearance: every figure has a full body, hidden sides match the visible design. Retain the original brushwork, palette, light and faces.
FIGURES: {CENTRAL_FIGURE: position, hair, clothing, hands}. {OTHER_FIGURES: left/right position, one line each}. Keep every pose, gesture, gaze and costume exactly. No figure moves.
ROOM: {CEILING}, {BACK_WALL_AND_WINDOWS}, {OBJECTS_ON_TABLE_OR_GROUND}.
0-3 s: hold the exact image about half a second, then fly forward and pass macro-close over {OBJECT_1}.
3-7 s: glide low to {OBJECT_2} and curve around it at macro distance.
7-10 s: rise along {HAND_OR_FABRIC} toward the face.
10-13 s: close on the central face; ease the speed for about a second; it does not move.
13-17 s: orbit behind the central figure toward {BACKGROUND_DETAIL}, then back over the other shoulder.
17-20 s: rise and retreat onto the opening axis and decelerate into the exact opening framing so the loop restarts.
Natural motion blur, shallow focus on macro passes, stable geometry and light, brushwork and canvas texture at every scale. No cuts, crash zooms, dissolves, flat canvas flyovers, scene morphing, extra people, new props, text or watermark.

Alive portrait: subject or camera channel (any)

Old family photo or portrait, 4-6 s of small motion; pick the subject channel (breath, blink) or the camera channel (push-in, everyone still). Adapted from Runware docs, Directing motion in image-to-video prompts (Runway Gen-4.5).

SUBJECT CHANNEL (a restored family photo or portrait, 4-6 s):
{SUBJECT} breathes slowly; the chest rises and falls gently. One natural blink at about {BLINK_SECOND} s. A very slight {HEAD_TILT or SOFTENING_OF_THE_SMILE}. {HAIR_OR_FABRIC} moves faintly in a soft breeze. {ATMOSPHERE: e.g. dust motes drift in the window light}. The eyes stay on {GAZE_TARGET}. The camera holds completely still. No change of pose, no speaking, no new people or objects. Otherwise the composition holds.

CAMERA CHANNEL (a group photo or a painting):
Slow continuous push-in toward {FOCAL_POINT}. Everyone remains completely still: no breathing, no blinking. Only the camera moves forward across the {SECONDS} seconds. {ATMOSPHERE}. Otherwise the composition holds.

Rule: if a sentence describes something already visible in the photo, delete it. Only write what moves, how fast, and what must not move.

Artwork characters fight (game HUD) (seedance_2_5)

15 s round where two public-domain painting figures fight with fighting-game HUD; make the opening still first (gpt_image_2_5), then a round 2 from @video1. Adapted from Ivanna Humeniuk on X, Mona Lisa vs Venus round 1 prompt (May 2026).

Use image1 as the opening frame: {FIGURE_A} left, {FIGURE_B} right. Full fighting-game HUD: yellow health bar left, green right, portraits, names, ROUND 1 center, SUPER bars bottom. Arena: {ARENA_MIXING_BOTH_ARTWORKS}. Photorealistic cinematic, classic 2D fighting-game staging. The HUD reacts to every action.
CHARACTERS: {FIGURE_A} (left): {LOOK_FROM_THE_ARTWORK}, uses {POWER_TIED_TO_THEIR_ARTWORK}. {FIGURE_B} (right): {LOOK}, uses {POWER}.
SHOT 1 (00:00-00:01): "ROUND 1" in gold text. Voice: "ROUND ONE. FIGHT!" Camera rockets in.
SHOT 2 (00:01-00:04): DIRECT COMBAT. {A_ATTACK}. {B_BLOCK}. {B_COUNTER}. 360 degree orbit. Slow motion 20% on impacts, {MATERIAL_A} and {MATERIAL_B} spraying outward.
SHOT 3 (00:04-00:07): SPECIAL ABILITIES. {A_SPECIAL}. {B_SPECIAL}. Extreme close-up as the two powers clash. Bars dropping.
SHOT 4 (00:07-00:10): TITAN CLASH. {A_GIANT_FORM} against {B_GIANT_FORM}. Slow motion 15%: massive collision center screen, camera rushes into the explosion.
SHOT 5 (00:10-00:15): {ENDING}. Both slammed into opposite pillars, slow orbit, both bars empty. FREEZE FRAME. "{END_TEXT}" burns center. Hard cut to black.

Trending formats to remake

To remake one: download the example with the media-download skill, take a frame a second and the transcript, write a beat list (time, shot, camera, on-screen text, sound), keep what the format needs, swap in the person's product or character (never the creator's face, voice, logo or footage), show the beat list and stills before spending, then make it with the skill named below.

  1. Step inside the painting: One continuous camera move from a famous painting into its 3D space (macro on objects, orbit behind a figure) and back to the exact original frame, so it loops. Why it spreads: It turns a familiar image into a discovery. 321 bookmarks on 40.8k views with the full prompt shared; the same creator's second painting fly-through also passed 300 bookmarks. Keep: The painting as the exact first and last frame, one continuous time-coded camera path (macro over objects, the central face, an orbit behind the figure, back out), figures frozen, brushwork kept at every scale, 20 s 9:16. Swap the painting (public domain). Make it with Image to Video (inference-sh-image-to-video).
  2. Artwork characters come alive: Figures from famous paintings step out and do something new: fight, dance or meet each other, while keeping the painted look. Why it spreads: Mona Lisa vs Venus as a fighting game reached 363k views and 1,979 bookmarks, with the prompts in the thread. Public-domain art gives instant recognition without likeness issues. Keep: Two public-domain artwork figures with their painted look, powers that come from their artworks, a game format everyone knows (health bars, ROUND 1, EPIC DRAW), 15 s of timed shots, round 2 continued from the first video. Swap the artworks and the powers. Make it with varg AI (varg-ai).
  3. Old family photo comes alive: Restore and colorize an old family photo, then add a few seconds of subtle motion, shown before and after. Why it spreads: It is the evergreen personal use: high save intent (an old-photo restore tutorial got 1,493 bookmarks on 95.6k views) and now built into Google Photos and TikTok. The best results live in private family chats, so we found no public post that passes the bar. Keep: Before and after, a light restore checked face by face against the original, 4-6 s of breath and one blink with the camera still, and an AI label. Swap the photo; never make the person speak. Make it with varg AI (varg-ai).

Open-source repos

Set up varg

varg is one API key for the models these skills drive — Seedance 2.5, Kling 4, Veo 3.1, Wan 3, Nano Banana 2.1, GPT Image 2.5, ElevenLabs v4 and 200+ more — billed in credits, with a cost estimate before anything runs.

Related tutorials