Long Videos with Veo 3.1 Extend

Veo 3.1 has a game-changing feature: video extension. Take any short clip and extend it up to 30 seconds total. Each extension adds 7 seconds of coherent continuation.

How It Works#

• Start with a base video (5-10 seconds)
• Feed it to Veo 3.1 extend
• Get 7 more seconds of continuation
• Repeat until you hit 30 seconds

The model maintains character consistency, motion direction, and scene coherence across extensions.

Ice Skating Princess - Extended#

Started with a short clip of an ice skating princess. Extended it multiple times to create a longer, flowing performance.

The model picked up on:

• The spinning motion and continued it naturally
• Costume details stayed consistent
• Ice surface reflections maintained
• Camera angle preserved

Tips for Clean Extensions#

Motion direction matters: If your character is moving left, the extension continues that direction. Plan your initial clip with the full sequence in mind.

Avoid cuts: End your base clip mid-motion, not at a natural stopping point. The model extends better from continuous movement.

Consistent lighting: Dramatic lighting changes confuse the extension. Keep your source evenly lit.

Simple backgrounds: Complex backgrounds introduce more opportunities for artifacts. Clean environments extend cleaner.

The Math#

• Base video: 5-10 seconds
• Each extension: +7 seconds
• Maximum total: 30 seconds
• Extensions needed for max: 3-4 rounds

When to Use This#

• Character performances that need length
• Establishing shots that feel too short
• Dance sequences
• Walking/running cycles
• Any continuous motion

When NOT to Use This#

• Scene changes mid-video
• Dialogue (lip sync doesn't extend well)
• Complex multi-character interactions
• Videos that need precise timing

Combining with Other Tools#

My workflow:

• Nano Banana Pro - Generate initial frame
• Kling 2.6 - Create 5-second base video
• Veo 3.1 Extend - Stretch to 20-30 seconds
• Topaz - Final upscale

The ice skating video above went through this exact pipeline. The result is a smooth 30-second performance from a single generated image.