8 min read
How to Turn AI Photos into Videos on the Canvas
Every image on the canvas is one click away from motion. Animate a finished photo into a short clip, direct the camera with a motion chip, extend the clip when it ends too soon, stitch several takes into one sequence — and when a moving frame is better than the photo you started with, screenshot it back onto the board as a new image. This tutorial covers the full video loop, from first animation to finished sequence.
step by step
Animate an image
Hover a completed image and choose Animate into Video (or use /video in chat). Describe the motion — "slow zoom in, hair blowing in the wind, she smiles" — pick a length between about 3 and 10 seconds, and the clip renders right beside its source photo.
Direct the camera
Add a camera chip from the rail to control how the virtual camera moves: presets like Cinematic Push-In, Handheld Follow, Orbit Reveal, Drone Pullback, Locked Off and Crash Zoom, or your own combination of move, rig and speed.
Extend and stitch
Use Extend on a video node to add 2–10 more seconds — describe what happens next, like "the camera slowly pulls back". Select multiple clips and use /stitch to merge up to five of them, in selection order, into one video.
Screenshot the best frame
Pause a video on a great moment and hit the camera icon in its toolbar. The frame becomes a new image node beside the video — ready to edit, use as a reference, or animate again in a new direction.
Writing motion prompts that work
Good video prompts describe change over time, not scene contents — the source image already supplies the scene. Name one or two subject actions ("she turns toward the camera and laughs") and at most one camera behavior ("slow push-in"). Overloaded motion prompts produce mush; a clip with one clear beat almost always lands. If you need three beats, that's three clips — generate, extend, extend.
The camera chip, briefly
The camera chip is the video counterpart to shot settings, and it speaks real cinematography.
Moves
Pan, tilt, dolly in and out, truck, boom, orbit, crane and tracking — plus zooms from slow creep to crash.
Rigs and speeds
Tripod, handheld, steadicam, gimbal, shoulder, drone and FPV each carry a distinct feel; speed runs from very slow to whip. Handheld at moderate speed reads as documentary; locked-off tripod reads as composed.
Signature moves
Dolly-zoom vertigo, whip pan, rack focus, 360° orbit and parallax reveal — the recognizable film-school moves, one click each.
Beyond animation: the rest of the video toolbox
The canvas also handles the practical parts. Trim a clip's ends, mute its audio, or overlay caption text at the top, center or bottom. Dialogue nodes hold reusable spoken lines for talking videos. Text-to-video works with no source image at all when you want motion from scratch. And on Pro and Studio plans you can restyle an existing clip generatively — "make this anime", "add falling snow", "black and white" — or switch between video providers per shot.
The image-to-video-to-image loop
The screenshot tool turns video into a photography session. Motion produces expressions and in-between moments that a still generator rarely gives you — a genuine mid-laugh, a coat caught in the wind. Capture the frame, and it's a first-class image node: edit it, download it, or feed it back as a reference for future stills. Some of the best "photos" on a canvas were never generated as photos at all.
quick tips
- Animate from your best still rather than describing a scene from scratch — the image pins character, outfit and lighting, so the video only has to solve motion.
- Match the camera to the mood: handheld follow for candid feed energy, slow push-in for drama, drone pullback for a big finish.
- Stitch order is selection order — click clips in the sequence you want before running /stitch.
- Extending an existing clip keeps continuity better than generating a fresh clip of "the next moment".
questions, answered
How long can a video be?+
Single generations run up to about 10 seconds, extensions add 2–10 seconds each, and stitching combines up to five clips — so a sequence of a minute or more is a stitch away.
Does video cost more than images?+
Video is billed by length (credits per second), so a 3-second test costs a fraction of a 10-second take. Iterate short, then extend the winner.
Will my character's face stay consistent in motion?+
Yes — animating a canvas image carries the face straight from the still into the clip. It's the most reliable route to consistent video.
Can I add speech?+
Yes — create dialogue nodes for spoken lines and use them in talking videos. Overlay text handles captions for sound-off viewing.
keep learning
shot settings →
Framing, lenses, lighting, color grades and camera styles: use selfielab's shot settings chip and presets like iPhone Candid or Studio Editorial to shoot like a photographer.
editing photos →
Relight a scene, swap the background, recolor an outfit or add a smile — selfielab's edit tools change one thing and keep the same face. Plus background removal and regenerate.
getting started →
Learn the selfielab canvas in five minutes: drop a character onto the infinite board, chat with the agent, and generate your first AI photo — same face, any scene.