| name | veo-extend |
| description | Generate character-consistent videos using Veo's video extension chain. Instead of generating separate clips and stitching, this creates one continuous video by extending each clip from the previous one's last frame. Solves character drift (especially for human characters like Mowgli) and eliminates stitching/sync issues. Use when generating multi-clip educational videos, when character consistency matters, or when the user mentions "extend", "chain clips", "consistent characters", or "continuous video". |
Veo Extension Pipeline
Generate perfectly synced, character-consistent videos by chaining Veo clip extensions. Each clip starts from the last frame of the previous one — characters, lighting, and style carry forward automatically.
Why Use This Instead of Separate Clips
| Problem with separate clips | How extension fixes it |
|---|
| Characters look different in each clip | Each extension starts from the last frame — same character carries forward |
| Stitching creates hard cuts or crossfade artifacts | One continuous video, no stitching needed |
| VO sync requires complex clip-plan math | Continuous video = just overlay full-vo.mp3 on top |
| Copyright/safety filters block image-to-video for human characters | First clip uses text-to-video, all extensions preserve that character |
How It Works
Frame → Clip 1 (8s) → Extend → Clip 1+2 (15s) → Extend → Clip 1+2+3 (22s) → ... → Full Video
Key facts:
- Initial clip: 4, 6, or 8 seconds
- Each extension: exactly 7 seconds (fixed)
- Maximum: 20 extensions = 148 seconds total (8s + 20×7s)
- Input: must be a Veo-generated video (MP4, 24fps, 720p/1080p)
Script Reference
Location: __PLUGIN_DIR__/scripts/extend-video.py
Initial generation (first clip):
python3 __PLUGIN_DIR__/scripts/extend-video.py \
--prompt "[Scene description] 3D animated cartoon. NO TEXT." \
--audio-prompt "ambient sounds, SFX description" \
--duration 8 --aspect "16:9" \
--output clip-chain-01.mp4
Extension (each subsequent clip):
python3 __PLUGIN_DIR__/scripts/extend-video.py \
--video clip-chain-01.mp4 \
--prompt "[NEW scene description — what happens next]" \
--audio-prompt "[NEW ambient sounds for this scene]" \
--output clip-chain-02.mp4
Each output contains the FULL video up to that point (not just the new 7s).
Final compositing:
ffmpeg -y \
-i clip-chain-final.mp4 \
-i audio/full-vo.mp3 \
-filter_complex "[0:a]volume=0.35[veo];[1:a]volume=1.0[vo];[veo][vo]amix=inputs=2:duration=shortest:normalize=0[aout]" \
-map 0:v -map "[aout]" \
-c:v copy -c:a aac -b:a 128k \
-t {VO_DURATION} \
final-video.mp4
Pipeline Steps
Step 1: Prepare prompts from timeline
Read audio/timeline.json. Map each clip to an extension prompt. Since extensions are fixed at 7s, regroup timeline clips into 7s windows.
Step 2: Generate initial clip
Pick the first scene. If human characters, use text-to-video. If animals only, use image-to-video.
Step 3: Chain extensions
For each subsequent 7s segment:
- Wait 60-90s between extensions (rate limit)
- Use previous output as
--video input
- Write a new prompt describing what happens in THIS 7s window
- Keep the same art style descriptors in every prompt
Step 4: Overlay VO
ffmpeg -y -i final-chain.mp4 -i audio/full-vo.mp3 \
-filter_complex "[0:a]volume=0.35[veo];[1:a]volume=1.0[vo];[veo][vo]amix=inputs=2:duration=shortest:normalize=0[aout]" \
-map 0:v -map "[aout]" -c:v copy -c:a aac -t {TOTAL_VO_DURATION} \
final-with-vo.mp4
Limitations
- Extension is always 7s — cannot request 4s or 6s extensions
- Only Veo-generated videos can be extended
- Cannot use reference images with extensions
- Videos expire from Google servers after 2 days — always save locally
- Max 148s total
- Scene changes are gradual — write prompts for natural visual transitions
Cost
- Initial clip (8s fast): ₹100
- Each extension (7s fast): ₹88
- 20 extensions for 148s video: ₹1,860
- Equivalent separate clips: ~₹3,000+
