Midjourney video uses an image as the starting frame. Current controls and GPU costs can change, so check the Video docs before production work.
1. Choose a strong starting frame
The starting image should already have the subject, composition and visual style you want. Video generation is not a substitute for fixing a bad base image.
2. Choose manual animation for control
Use the manual animation path when you want to write or adjust the motion prompt instead of letting the system decide automatically.
3. Pick motion amount deliberately
Use lower motion for subtle subject/camera movement and higher motion when larger movement is important, while watching for instability.
4. Describe motion, not the static image
Tell the model what changes over time: subject movement, wind, water, camera push, orbit, pan or other motion.
5. Extend or loop only after the first segment works
Fix identity and motion problems first. Then use extension/loop/end-frame controls when the current interface supports the sequence you need.
Animate this starting frame with [low/high] motion. The subject [action]. Secondary motion: [environment]. Camera [movement]. Preserve face/wardrobe/object layout. No new objects or scene change.
Trying to repair the starting image during animation; using high motion by default; adding new characters mid-shot; extending a flawed first segment and multiplying the error.
Finish check
The first short clip should already have stable identity and believable motion before you spend more GPU time extending it.