Loading
Production Guides
Studio-grade notes for live visuals, stage screens, and motion tests before the work moves into playback, compositing, or finishing.
Read This First
This guide is written for the MotionLab video model. MotionLab starts from an image and turns it into a moving clip of 4–15 seconds — so your prompt's job is to direct what happens over time, not to repaint the picture. Renders now carry a soundtrack too, but it is generated from what happens on screen, so keep every sentence about what the camera can see.
Pro Tip: If you re-describe the image in detail, the model spends its effort confirming the frame instead of animating it. One short anchor sentence about the scene is enough — then move straight to motion.
The Golden Rule
The model already sees your image. Focus every sentence on what happens next: how the subject moves, how the camera travels, and what shifts in the light and atmosphere around them.
[Scene anchor] + [Subject action] + [Camera movement] + [Environment in motion]
"A red sports car parked on a wet city street at night, neon signs, reflections, cinematic, 4K."
"The camera slowly orbits the parked sports car as neon reflections slide across its wet paint. Steam drifts up from a grate behind it, and light rain begins to streak through the glow of the signs."
Writing the Prompt
Write one flowing paragraph in present tense — not keyword tags, not bullet points. Use connectors like "as," "then," and "while" so actions flow into each other naturally from beginning to end.
"The dancer spins, her dress flaring outward" — not "dancer, spinning, dramatic, HD." The model understands natural language and the relationships between elements.
2–3 sentences of motion suit a 4–6 second clip; a 12–15 second clip needs 5–8 sentences describing a sequence of beats. A short prompt on a long clip makes the model rush or drift.
Give longer clips an order of events: "The camera holds steady at first. Then it begins a slow push-in as the fog thickens and the figure turns toward the light."
Write "her shoulders slump and she stares at the empty cup" instead of "she is sad." The model renders what a camera can see, not internal states.
Cinematography
The model understands standard cinematography vocabulary. Commit to one clear camera behavior per clip — a single clean move always looks better than two competing ones.
Dolly in / pull back, tracking shot alongside the subject, pan across, tilt upward, slow orbit, handheld follow, or a static locked-off frame.
Pair the move with a speed: "subtle," "gentle," "steady," "gradual" for calm shots — "sweeping," "rapid" for energy. An unqualified camera move tends to come out medium-fast.
Close-up, medium shot, wide establishing shot, overhead view. Close-ups need more fine detail in the prompt; wide shots need more environment.
Film grain, lens flares, shallow depth of field, motion blur, slow motion, time-lapse — name them explicitly if you want them.
Setting the Mood
Light and atmosphere are the strongest mood levers — and the best source of motion in otherwise static parts of the frame.
Pick a single light source and give it behavior: "low golden-hour sun casts long shadows," "a neon sign flickers, tinting the fog pink." Conflicting light directions confuse the model.
Fog rolls, dust drifts through light beams, rain streaks, reflections shift, embers rise. Atmosphere in motion is what makes a clip feel alive.
If part of your image is empty — sky, water, a wall — assign it motion: "clouds drift slowly across the sky," "ripples spread across the still water."
"Watercolor," "hand-drawn animation," "claymation," "analog film" — committing to a medium keeps the style stable across every frame.
Bring Your Own Movement
Drop an MP4 into the cyan MOTION slot next to your image and the render follows that clip's movement instead of inventing its own. It transfers motion — the camera path, the timing, the rhythm — not the look: the subject, palette, and style still come from your start image and prompt.
A move you already storyboarded, a beat you have to hit, a camera path from an existing shot, or a gesture you want repeated across several clips. A phone clip is enough — it is read for movement, not quality.
MP4, up to 50 MB and 60 seconds, and at least as long as the clip you are rendering. MotionLab re-encodes it in your browser to the fps you selected and trims it to your duration before it is sent.
Keep writing the shot as usual. The reference decides how things move; the prompt decides what is moving, how it is lit, and what the atmosphere does.
A reference video adds a fixed amount of credits to the render, shown in the price next to the generate button before you commit.
VJ & Motion Graphics
Turn on Loop in the video page and MotionLab renders the clip back to its opening frame — a seamless loop straight out of the generator, with no crossfade pass in an editor. Loop runs on MotionLab image-to-video renders: it needs a start image (the frame it returns to) and cannot be combined with an End image. Every Loop render is seam-checked, and the clip is marked when the join is still visible.
The switch closes the seam; the motion decides whether it reads. Direction like "cyclic motion," "steady rhythm," or "continuous repetitive movement" — wind, waves, smoke, rotation, breathing light — loops convincingly. A one-time event inside the clip still reads as a restart on the second pass.
The soundtrack is joined smoothly at the seam. If you upload your own audio, it only loops cleanly when its length matches the clip — so you can either keep the smooth join or mute the audio and let the video loop on its own.
Keep the background description minimal or solid-colored if you plan to extract the subject later for transparent overlays.
Audio
MotionLab renders used to be silent. They are not any more: the Audio switch on the video page is on by default, and the soundtrack is generated to match what happens on screen. You can also hand the render a track of your own, or write music from scratch in the separate Music mode.
Pro Tip: For lyrics, use tags such as [Verse], [Chorus], and [Bridge]; for a timed arrangement, write lines like "[0:00 - 0:10] soft piano intro". Choosing Instrumental explicitly is more reliable than hoping a prompt stays wordless.
On by default — the finished MP4 carries a generated soundtrack. Turn it off when you need a silent deliverable: LED walls, loops running under a live set, or footage going into an edit with its own mix.
Drop an MP3 or WAV into the amber AUDIO slot and the render follows your track's timing and energy instead of generating its own. The file is trimmed in your browser to the clip length before upload, and the render costs a small add-on.
With Loop on, the audio is joined smoothly at the seam. If you uploaded your own track, it only loops cleanly when its length matches the clip — so you can keep the smooth join, or mute the audio and let the picture loop on its own.
Switch the create dock to Music to generate a track: Lyria 3 writes 30 seconds, Lyria 3 Pro up to three minutes, both as MP3. Describe the brief in plain language, then open Advanced for genre, mood, instruments, structure, BPM, key, vocals, language, lyrics, and a timed arrangement.
Strengths
The MotionLab model excels at these types of content:
Wide, medium, and close-up shots with thoughtful lighting, shallow depth of field, and natural motion.
Strong single-subject emotional expressions, subtle gestures, and facial nuance.
Fog, mist, golden-hour light, rain, reflections, ambient textures.
Painterly, noir, analog film, fashion editorial, pixelated animation, stop-motion, claymation.
Known Limitations
To get the best results, avoid these patterns:
Chat-to-Motion-Graphics
Motion Studio is a different kind of tool: instead of generating footage from a model, you describe a motion-graphics piece in chat — a stage opener, kinetic type, an LED loop, a product reveal — and it is composed as live motion code that previews instantly and exports to MP4 from your browser. It works best for graphic, typographic, and geometric motion rather than photorealistic scenes.
Pro Tip: Motion Studio is in beta: behavior and available models may change, and complex requests can take a retry or two. Precise, incremental direction gets the best results.