Video Generation Skill
Purpose
Generate high-quality video content through structured, reproducible pipelines.
Pipeline Stages
- Concept: Creative brief, target audience, platform
- Pre-visualization: Storyboard, keyframes, style references
- Image Generation: Lock first frames with consistent style
- Video Generation: Animate using image-to-video
- Post-Processing: Upscale, color grade, assemble
- Audio: Voiceover, music, SFX, mixing
- Export: Format, codec, resolution, platform specs
Generation Strategy
| Approach |
Best For |
Tools |
| Text-to-video |
Exploration, prototyping |
Kling, Runway, Sora |
| Image-to-video |
Quality, consistency |
Kling, Runway Gen-4 |
| Video-to-video |
Style transfer, editing |
Runway, Seedance |
CF-Specific Flow
# 1. Generate/Load keyframe
keyframe = await generate_keyframe(prompt, style_refs)
# 2. Generate video clip
clip = await generate_video(
image=keyframe,
motion_prompt="slow dolly in, gentle camera pan",
duration=5, # seconds
fps=24,
)
# 3. Post-process
clip = await upscale(clip, target_resolution="1080p")
clip = await color_grade(clip, style="cinematic_warm")
# 4. Add audio
audio = await generate_voiceover(script, voice_id)
audio = await mix_audio(audio, music_track, sfx)
# 5. Assemble
video = await assemble_clips([clip1, clip2, ...], audio)
Quality Checklist
Cost Optimization
- Batch similar scenes together
- Use image-to-video over text-to-video
- Generate short clips (2-6s), assemble in post
- Use self-hosted models (Wan 2.2) for volume
- Pre-warm GPU pods for batch jobs
- Cache intermediate assets (keyframes, voice)
Consistency
- Character identity:
utils.pipeline.character_identity
- Voice:
utils.pipeline.voice_consistency
- Style: Seed management, reference anchoring
- Brand:
utils.pipeline.brand_memory
Background music mix
- Royalty-free sources OK when licensed; keep voiceover clear (duck music under VO)
- FFmpeg mix: volume, fade in/out; CF already has
mix_audio / ACE-Step music — prefer the CF pipeline audio stage when available
- License gate before publish (confirm rights for every track)
- Acceptance:
@cf-video-loop checklist covers audio sync
1---2name: ivx-cf-video-generation3description: Design and execute video generation pipelines using diffusion and transformer models. Use when generating video content, optimizing video quality, or building video production workflows.4---5# Video Generation Skill67## Purpose89Generate high-quality video content through structured, reproducible pipelines.1011## Pipeline Stages12131. **Concept**: Creative brief, target audience, platform142. **Pre-visualization**: Storyboard, keyframes, style references153. **Image Generation**: Lock first frames with consistent style164. **Video Generation**: Animate using image-to-video175. **Post-Processing**: Upscale, color grade, assemble186. **Audio**: Voiceover, music, SFX, mixing197. **Export**: Format, codec, resolution, platform specs2021## Generation Strategy2223| Approach | Best For | Tools |24|----------|---------|-------|25| Text-to-video | Exploration, prototyping | Kling, Runway, Sora |26| Image-to-video | Quality, consistency | Kling, Runway Gen-4 |27| Video-to-video | Style transfer, editing | Runway, Seedance |2829### CF-Specific Flow3031```python32# 1. Generate/Load keyframe33keyframe = await generate_keyframe(prompt, style_refs)3435# 2. Generate video clip36clip = await generate_video(37 image=keyframe,38 motion_prompt="slow dolly in, gentle camera pan",39 duration=5, # seconds40 fps=24,41)4243# 3. Post-process44clip = await upscale(clip, target_resolution="1080p")45clip = await color_grade(clip, style="cinematic_warm")4647# 4. Add audio48audio = await generate_voiceover(script, voice_id)49audio = await mix_audio(audio, music_track, sfx)5051# 5. Assemble52video = await assemble_clips([clip1, clip2, ...], audio)53```5455## Quality Checklist5657- [ ] First frame matches reference58- [ ] Motion is smooth (no flicker/judder)59- [ ] Subject consistency across frames60- [ ] No artifacts or distortions61- [ ] Resolution meets target (1080p minimum)62- [ ] Frame rate consistent (24/30/60 fps)63- [ ] Audio synced to video64- [ ] Color grade consistent across clips6566## Cost Optimization6768- Batch similar scenes together69- Use image-to-video over text-to-video70- Generate short clips (2-6s), assemble in post71- Use self-hosted models (Wan 2.2) for volume72- Pre-warm GPU pods for batch jobs73- Cache intermediate assets (keyframes, voice)7475## Consistency7677- Character identity: `utils.pipeline.character_identity`78- Voice: `utils.pipeline.voice_consistency`79- Style: Seed management, reference anchoring80- Brand: `utils.pipeline.brand_memory`8182## Background music mix8384- Royalty-free sources OK when licensed; keep voiceover clear (duck music under VO)85- FFmpeg mix: volume, fade in/out; CF already has `mix_audio` / ACE-Step music — prefer the CF pipeline audio stage when available86- License gate before publish (confirm rights for every track)87- Acceptance: `@cf-video-loop` checklist covers audio sync
intelli-verse-x