Video Prompts
Specify the subject, camera movement (slow pan, aerial orbit, tracking shot), lighting, duration (4s, 6s, 8s), and aspect ratio (16:9 widescreen or 9:16 vertical).
Swarming
Video, audio, and studio
Swarm brings generative video, image motion, sound design, and video editing into a unified creative workflow. Ask for videos, images, and audio clips directly in chat, chain keyframes across multi-scene stories, remix visuals, and edit your timeline in Video Studio with real-time AI sidecar guidance.
Media quick settings
/media keeps setup close to the conversation.
In Desktop, enter /media to open Media Quick Settings without leaving chat. Use it to register or review source video folders, inspect configured media models, and check background transcription status.
Provider support: Google’s media generation models are currently supported under the hood as our initial backend. We are actively working to branch out to additional providers as soon as time permits. All prompt patterns, tools, and Desktop interfaces are designed to be model-neutral.
Quick start
In Desktop, enter /media to verify your media settings. Then return to chat and ask Swarm for the video, image, or sound clip you need.
Asking for media
How to ask for videos, images, and sound clips.
You can request media directly in chat just like code or text. To get the best results, describe the subject, style, mood, and technical specifications you need:
Image Prompts
Describe the subject, composition, atmosphere, color palette, and negative space, along with any elements to exclude.
Audio Prompts
State whether you want a sound effect, ambient bed, or musical theme, along with the desired duration (15s, 30s, or longer), mood, and instrumentation.
Quality Tiers
Ask for preview quality (360p/720p) for fast iterations, or high/master quality (1080p, 4K) when rendering final production assets.
Example video request
“Create a 6-second cinematic 16:9 video: an autonomous delivery drone descending gently through evening mist onto a rooftop landing pad, 35mm lens, soft orange beacon lights, and wet concrete reflections.”
Frame to video
Take an image frame or video still and turn it into motion.
You do not have to start video generation from scratch. Swarm allows you to take any visual still—a generated image, an extracted video frame, a UI mockup, or a vector graphic—and use it as the starting frame for a video clip.
Choose or Extract
Select an approved image artifact, or ask Swarm to extract a specific frame (first, last, or at an exact timestamp) from an existing video.
Direct the Motion
Tell Swarm how the camera or subjects should move while preserving character, environment, and lighting continuity.
Generate Video
Swarm seeds the video model with the exact frame pixels, producing a natural motion clip that seamlessly starts where the image left off.
SVG Auto-Rasterization
Passing a vector SVG logo or icon? Swarm automatically rasterizes it to a crisp high-res PNG via headless Chromium before generating motion.
Example image-to-video prompt
“Take this workstation still we just approved. Keep the composition and lighting identical, and generate a 6-second video showing gentle steam rising from the coffee mug with a slow subtle push-in toward the screen.”
Multi-part videos
How to ask for multi-scene stories with keyframe chaining.
Single AI video clips are bounded in duration. To produce longer continuous stories without jarring jumps between scenes, ask Swarm for a multi-part video. Swarm plans each scene, chains keyframes across boundaries, and stitches the clips together.
Structured Story
Swarm breaks your prompt into sequential scenes with defined actions, camera transitions, and visual focus points.
Frame Continuity
Swarm automatically extracts the concluding frame of Scene 1 and uses it as the visual seed for Scene 2, ensuring seamless continuity.
Cuts & Crossfades
Connect consecutive parts with precise hard cuts or smooth crossfades on the timeline.
Video Stories
Request 2 to 16 scenes in one turn. Swarm generates the clips, keyframes, transitions, and soundtrack in a single orchestrated operation.
Example multi-part video prompt
“Create a 3-part video story for a deep-sea exploration submersible:
• Part 1: The sub descends from turquoise surface water into dark midnight depths with its exterior hull lights flashing.
• Part 2: Submersible lights illuminate a glowing hydrothermal vent on the ocean floor.
• Part 3: A rare bioluminescent creature swims gracefully past the viewport as the sub steadies its position.”
Image remixes
Iterate on existing visuals with preserve, change, and exclude.
When you have a promising image or visual design, you don’t want to regenerate from scratch and lose its character. Ask Swarm to remix the exact artifact, specifying what to keep and what to modify.
Lock Key Elements
Explicitly name the subject, character, camera angle, product geometry, or composition that must remain identical.
State the Edit
Describe the targeted alteration: shifting time of day, changing background environment, tweaking colors, or adjusting aspect ratio.
Prevent Drift
List details you want to prevent: extra people, unwanted text or logos, or changes to locked elements.
Example image remix prompt
“Remix this research laboratory still: preserve the central console, scientist silhouette, and camera angle. Change the neutral white lighting to emergency amber strobes with warning telemetry holograms floating above the screens. Do not add extra people.”
Sound clips & music
Generate sound effects, ambient beds, and original songs.
Audio brings visuals to life. You can ask Swarm for standalone sound effects, ambient loops, or full musical compositions of specific durations.
- Flexible durations: Request short sound effects (15s, 30s) or extended musical themes up to several minutes.
- Multi-variation prompts: Ask for multiple musical or sound takes in a single turn (e.g. ambient synth, orchestral, lo-fi) to audition different arrangements.
- Interactive sound bars: Desktop provides interactive audio player bars with instant waveform scrubbing, playback controls, and volume adjustment.
- Sound library management: Audio files are registered as durable artifacts you can preview, download, or drop into any Video Studio project.
Example sound clip request
“Generate a 30-second retro sci-fi soundscape: warm analog synthesizer pads, a gentle hypnotic arpeggio, subtle mechanical clicks, and tape saturation suitable for a technology demo.”
Sound in videos
Integrate soundtracks and Foley with automated audio ducking.
You can ask Swarm to pair any video project with custom audio. Swarm handles alignment, looping, and volume balancing automatically.
Continuous Music
Span a single continuous music track across multiple video scenes to maintain rhythmic and tonal unity throughout your story.
Automated Ducking
Background music automatically ducks in volume whenever dialogue or prominent sound effects occur, ensuring spoken words and Foley remain crisp.
Mix or Override
Choose whether generated music replaces native video audio entirely, mixes ducked, or layers alongside existing source tracks.
Beats & Transcripts
Swarm analyzes audio waveforms and transcription boundaries to synchronize clip cuts and visual transitions with musical beats.
Example audio integration prompt
“Take our 3-part submersible video and add a tense, atmospheric electronic soundtrack. Ensure the music ducks automatically whenever the thruster sounds or sonar pings play.”
Video Studio & sidecar
Make and edit videos with timeline controls and the AI sidecar.
Swarm Desktop includes a dedicated Video Studio surface alongside your active chat session. You can fluidly move between high-level creative prompts and direct timeline editing.
Context-Aware Chat
The sidecar chat automatically embeds the current playhead timestamp, selected clip ID, and proposal context so Swarm knows exactly what you are referencing.
Working Cuts
AI edits arrive as proposed working cuts with visual diff indicators. Compare variations side-by-side, test playback, and confirm the cut to commit an immutable revision (r0, r1, r2).
Canvas Edit Boxes
Manipulate split-screens, Picture-in-Picture (PiP), text overlays, and corner masks directly on the canvas using interactive drag-and-resize handles.
HTML Animation Clips
Convert native HTML animations or storyboard artifacts directly into timeline clips with synchronized audio playback without waiting for pre-rendering.
Background render center
When your timeline is ready, dispatch render jobs to the background FFmpeg render engine. Monitor real-time progress bars and export the final MP4 as an immutable managed artifact at preview, standard, high, or master quality.
Example Video Studio prompt
“Trim the first clip by 1.5 seconds, crossfade into the second scene over 500ms, and add a Picture-in-Picture window in the top right showing the drone telemetry feed.”
Sources & transcripts
Registered folders are workspace-scoped and read-only.
- Register only the folders you want Swarm to browse for the current workspace.
- Swarm navigates bounded subdirectories and works from opaque source references rather than arbitrary paths.
- Source access permits viewing and transcription; it never permits editing, moving, renaming, or deleting your original files.
- Transcription jobs run in the background, extracting word-timed speech, topic section indexes, and waveform buckets for audio alignment.
- Video understanding inspects visible frames and embedded audio tracks to locate clips and topics.
Use focus notes for silent demos
Tell Swarm when a clip is a silent software demo and ask for a dense play-by-play of cursor actions, navigation, visible text changes, and results. Focus notes guide analysis without distorting facts.
Providers & roadmap
Model-neutral workflows with multi-provider expansion underway.
Swarm's media architecture is designed to decouple your creative workflow and prompts from specific vendor APIs:
- Current support: Google’s media generation models are currently supported under the hood as our initial production backend.
- Roadmap expansion: We are actively branching out to support additional video, audio, and image providers as soon as development time allows.
- Direct provider billing: Generations utilize connected provider APIs and are billed directly to your provider accounts per second generated or per asset, with upfront estimates before execution.
- Regional policies: External video generation may be subject to regional provider availability rules (such as EU/EEA restrictions). Check your provider account settings when configuring media keys.