inference-sh/ai-content-pipeline

AI Content Pipeline

Build multi-step content creation pipelines via inference.sh CLI.

Quick Start

Requires inference.sh CLI (belt). Install instructions

belt login

# Simple pipeline: Generate image -> Animate to video
belt app run falai/flux-dev --input '{"prompt": "portrait of a woman smiling"}' > image.json
belt app run falai/wan-2-5 --input '{"image_url": "<url-from-previous>"}'

Pipeline Patterns

Pattern 1: Image -> Video -> Audio

[FLUX Image] -> [Wan 2.5 Video] -> [Foley Sound]

Pattern 2: Script -> Speech -> Avatar

[LLM Script] -> [Kokoro TTS] -> [OmniHuman Avatar]

Pattern 3: Research -> Content -> Distribution

[Tavily Search] -> [Claude Summary] -> [FLUX Visual] -> [Twitter Post]

Complete Workflows

YouTube Short Pipeline

Create a complete short-form video from a topic.

# 1. Generate script with Claude
belt app run openrouter/claude-sonnet-45 --input '{
  "prompt": "Write a 30-second script about the future of AI. Make it engaging and conversational. Just the script, no stage directions."
}' > script.json

# 2. Generate voiceover with Kokoro
belt app run infsh/kokoro-tts --input '{
  "prompt": "<script-text>",
  "voice": "af_sarah"
}' > voice.json

# 3. Generate background image with FLUX
belt app run falai/flux-dev --input '{
  "prompt": "Futuristic city skyline at sunset, cyberpunk aesthetic, 4K wallpaper"
}' > background.json

# 4. Animate image to video with Wan
belt app run falai/wan-2-5 --input '{
  "image_url": "<background-url>",
  "prompt": "slow camera pan across cityscape, subtle movement"
}' > video.json

# 5. Add captions (manually or with another tool)

# 6. Merge video with audio
belt app run infsh/media-merger --input '{
  "video_url": "<video-url>",
  "audio_url": "<voice-url>"
}'

Talking Head Video Pipeline

Create an AI avatar presenting content.

# 1. Write the script
belt app run openrouter/claude-sonnet-45 --input '{
  "prompt": "Write a 1-minute explainer script about quantum computing for beginners."
}' > script.json

# 2. Generate speech
belt app run infsh/kokoro-tts --input '{
  "prompt": "<script>",
  "voice": "am_michael"
}' > speech.json

# 3. Generate or use a portrait image
belt app run falai/flux-dev --input '{
  "prompt": "Professional headshot of a friendly tech presenter, neutral background, looking at camera"
}' > portrait.json

# 4. Create talking head video
belt app run bytedance/omnihuman-1-5 --input '{
  "image_url": "<portrait-url>",
  "audio_url": "<speech-url>"
}' > talking_head.json

Product Demo Pipeline

Create a product showcase video.

# 1. Generate product image
belt app run falai/flux-dev --input '{
  "prompt": "Sleek wireless earbuds on white surface, studio lighting, product photography"
}' > product.json

# 2. Animate product reveal
belt app run falai/wan-2-5 --input '{
  "image_url": "<product-url>",
  "prompt": "slow 360 rotation, smooth motion"
}' > product_video.json

# 3. Upscale video quality
belt app run falai/topaz-video-upscaler --input '{
  "video_url": "<product-video-url>"
}' > upscaled.json

# 4. Add background music
belt app run infsh/media-merger --input '{
  "video_url": "<upscaled-url>",
  "audio_url": "https://your-music.mp3",
  "audio_volume": 0.3
}'

Blog to Video Pipeline

Convert written content to video format.

# 1. Summarize blog post
belt app run openrouter/claude-haiku-45 --input '{
  "prompt": "Summarize this blog post into 5 key points for a video script: <blog-content>"
}' > summary.json

# 2. Generate images for each point
for i in 1 2 3 4 5; do
  belt app run falai/flux-dev --input "{
    \"prompt\": \"Visual representing point $i: <point-text>\"
  }" > "image_$i.json"
done

# 3. Animate each image
for i in 1 2 3 4 5; do
  belt app run falai/wan-2-5 --input "{
    \"image_url\": \"<image-$i-url>\"
  }" > "video_$i.json"
done

# 4. Generate voiceover
belt app run infsh/kokoro-tts --input '{
  "prompt": "<full-script>",
  "voice": "bf_emma"
}' > narration.json

# 5. Merge all clips
belt app run infsh/media-merger --input '{
  "videos": ["<video1>", "<video2>", "<video3>", "<video4>", "<video5>"],
  "audio_url": "<narration-url>",
  "transition": "crossfade"
}'

Pipeline Building Blocks

Content Generation

| Step | App | Purpose | |------|-----|---------| | Script | openrouter/claude-sonnet-45 | Write content | | Research | tavily/search-assistant | Gather information | | Summary | openrouter/claude-haiku-45 | Condense content |

Visual Assets

| Step | App | Purpose | |------|-----|---------| | Image | falai/flux-dev | Generate images | | Image | google/imagen-3 | Alternative image gen | | Upscale | falai/topaz-image-upscaler | Enhance quality |

Animation

| Step | App | Purpose | |------|-----|---------| | I2V | falai/wan-2-5 | Animate images | | T2V | google/veo-3-1-fast | Generate from text | | Avatar | bytedance/omnihuman-1-5 | Talking heads |

Audio

| Step | App | Purpose | |------|-----|---------| | TTS | infsh/kokoro-tts | Voice narration | | Music | infsh/ai-music | Background music | | Foley | infsh/hunyuanvideo-foley | Sound effects |

Post-Production

| Step | App | Purpose | |------|-----|---------| | Upscale | falai/topaz-video-upscaler | Enhance video | | Merge | infsh/media-merger | Combine media | | Caption | infsh/caption-video | Add subtitles |

Best Practices

1. Plan the pipeline first - Map out each step before running
2. Save intermediate results - Store outputs for iteration
3. Use appropriate quality - Fast models for drafts, quality for finals
4. Match resolutions - Keep consistent aspect ratios throughout
5. Test each step - Verify outputs before proceeding

Related Skills

# Video generation models
npx skills add inference-sh/skills@ai-video-generation

# Image generation
npx skills add inference-sh/skills@ai-image-generation

# Text-to-speech
npx skills add inference-sh/skills@text-to-speech

# LLM models for scripts
npx skills add inference-sh/skills@llm-models

# Full platform skill
npx skills add inference-sh/skills@infsh-cli

Browse all apps: belt app list

Documentation

• Content Pipeline Example - Official pipeline guide
• Building Workflows - Workflow best practices