Search or generate media from inside the native edit, place the returned operation deliberately, create captions, inspect audio, and combine frame, motion, and playblast review.
Where: Native Video Editor → AI, Media, Assets, or Captions.
Key ideas
- Find, make, and place are separate: Stock search or generation produces a candidate/operation. Placement chooses the active composition, frame, and editorial use. Existing project assets should be placed rather than regenerated.
- Paid media uses explicit authorization: Image, video, voice, music, SFX, and 3D tools expose catalog/quote information and retain their Sparks gates. An MCP host or assistant cannot treat a quote as permission to spend.
- Review has three evidence types: Rendered frames verify sampled composition; numerical motion review finds holds/jumps; a bounded 3D playblast shows continuous scene motion. Normal playback/export remains necessary for rhythm, sync, and audio quality.
Steps
- Open Assets to check whether the needed project source already exists; place it instead of making a duplicate.
- Open Media or ask AI to search stock for image, video, music, SFX, 3D model, or HDRI. Inspect the candidate before placement.
- For generation, state purpose, subject, duration/aspect/style, exact constraints, reference roles, and audio intent. Read the media catalog and quote first.
- Approve the exact quoted Sparks only when ready, generate, then inspect the saved operation receipt.
- Place the chosen operation at the intended frame/track. For 3D, generated GLBs enter as editable scene layers; place HDRIs into a World environment workflow.
- For speech, inspect available voices or existing audio and generate/replace only the required element.
- Use Captions on a supported audio source, review timing and text as native caption layers, and keep composition caption/motion-blur settings as the render source of truth.
- Run frame review at establishing, action, edit-boundary, reading-hold, and final frames; repair copy, crop, hierarchy, safe margins, or missing subjects.
- Run motion review over dense short ranges around edits/easing. For 3D, also make a short playblast and inspect camera motion, occlusion, action, easing, and loop seam.
- Play and export the complete piece to accept temporal rhythm, caption sync, voice/music balance, and continuous motion.
Media tools
- Search/place stock and project assets
- Media model catalog plus quote/generate/place operation
- Image, video, voice, music, and sound effects
- Voice catalog and bounded audio-range inspection
- Caption creation from supported audio
- 3D stock/HDRI, quoted model generation, and playblast
Evidence boundaries
- Frames: actual composition, copy, crop, hierarchy, safe margins, RTL, requested subjects
- Motion samples: activity, position, opacity, displacement, numerical discontinuities
- Playblast: continuous rendered 3D motion, silent and short
- Normal playback/export: edit rhythm, perceived easing, caption sync, audio quality and balance
Tips
- Review both sides of every important edit, not just representative middles.
- For loops/procedural motion, sample nonuniform normalized times such as .23, .41, .68, and .87 before the playblast.
- After two identical failures, change the authoring approach instead of repeating the same call.
Limitations and important notes
- Playblasts are silent WebM, limited to 10 seconds and 1–30 fps.
- A successful generation receipt does not mean the result was placed or creatively accepted.
- Caption and media services depend on authentication/provider availability.
- Still review cannot validate audio or continuous motion.
Troubleshooting
Generated media exists but is absent from the edit.
Read compositor media operations and place the ready operation into the active composition at the intended frame.
A review says frames passed but motion still feels wrong.
Frames are sampled evidence. Run motion review, play the range continuously, and inspect an export.
3D generation will not start.
Request a quote, confirm the quality/cost, and provide the exact authorization expected by the generation tool.