Model Chaining Skill

Overview

Create sequential AI model pipelines where output from one model feeds into the next, enabling complex multi-stage content generation.

Capabilities

create_chain

Build multi-model execution pipelines.

Inputs:

• name: Chain name
• description: What the chain produces
• steps: List of model steps with configs

Outputs:

• chain_id: Unique chain identifier
• model_count: Number of models in chain
• estimated_cost: API cost estimate

Example:

create_chain(
    name="Product Video Pipeline",
    steps=[
        {"model": "flux_pro", "type": "image", "prompt": "{{concept}}"},
        {"model": "runway_gen3", "type": "video", "image": "{{step_1}}"},
        {"model": "stable_audio", "type": "audio", "style": "upbeat"},
        {"model": "video_compositor", "video": "{{step_2}}", "audio": "{{step_3}}"}
    ]
)

execute_chain

Run chained model sequence with automatic data passing.

Inputs:

• chain_id: Chain to execute
• input_data: Initial input for first model
• parameters: Override default configs

Outputs:

• outputs: Result from each model step
• final_output: Final pipeline output
• artifacts: All intermediate results

chain_templates

Pre-built pipelines for common workflows.

Inputs:

• category: image/video/content/product
• complexity: simple/moderate/advanced

Outputs:

• templates: Available chain templates

Built-in Templates:

Image Enhancement Chain

1. Image generation (FLUX Pro)
2. Background removal
3. Upscaling (4x)
4. Style transfer

Video Production Chain

1. Script generation (LLM)
2. Storyboard images
3. Image-to-video
4. Music generation
5. Video editing/composition

Product Design Chain

1. Concept generation (text)
2. Design creation (image)
3. Multiple variations (editing)
4. Mockup generation
5. Product creation

Content Marketing Chain

1. Topic research (web search)
2. Article writing (LLM)
3. Feature image generation
4. Social graphics (image variants)
5. Video summary creation

step_reordering

Rearrange pipeline steps.

Inputs:

• chain_id: Chain to modify
• step_id: Step to move
• new_position: Target position

Outputs:

• updated_order: New step sequence

conditional_branching

Add logic gates to chains.

Inputs:

• chain_id: Chain to modify
• after_step: Step to add condition after
• condition: Evaluation expression
• true_branch: Steps if condition true
• false_branch: Steps if condition false

Outputs:

• branch_id: Branch identifier

Example:

conditional_branching(
    after_step=1,
    condition="{{quality_score}} > 0.85",
    true_branch=[{"model": "publish"}],
    false_branch=[{"model": "regenerate"}]
)

output_transformation

Format data between models.

Inputs:

• chain_id: Chain to modify
• between_steps: [source_step, target_step]
• transform: Transformation function

Outputs:

• transform_id: Transformation identifier

Transforms:

• extract_url: Get URL from model response
• resize_image: Resize image to specific dimensions
• extract_text: Pull text content from response
• format_json: Convert to JSON structure
• merge_outputs: Combine multiple step outputs

save_chain_template

Save custom chain as reusable template.

Inputs:

• chain_id: Chain to save
• template_name: Name for template
• category: Template category
• description: What it does

Outputs:

• template_id: Saved template identifier

parallel_execution

Run multiple models simultaneously.

Inputs:

• chain_id: Chain to modify
• steps: Steps to run in parallel
• merge_strategy: How to combine outputs

Outputs:

• parallel_group_id: Parallel execution group

Model Categories

Image Models

• flux_pro: Highest quality images
• flux_dev: Fast iterations
• sdxl: Stable Diffusion XL
• ideogram: Text rendering
• recraft_v3: Vector graphics

Video Models

• runway_gen3: Text/image to video
• kling_v2: High-quality video
• luma_dream_machine: Creative video
• pika: Video editing

Editing Models

• photomaker: Photo editing
• controlnet: Structured editing
• instruct_pix2pix: Instruction-based editing

Audio Models

• stable_audio: Music generation
• eleven_labs: Text-to-speech
• musicgen: Background music

3D Models

• shap_e: 3D generation
• wonder3d: Image to 3D

Text Models

• llama_70b: Text generation
• gpt4: Advanced reasoning
• mixtral: Fast text

Chain Patterns

Sequential Pipeline

Model 1 → Model 2 → Model 3 → Output

Parallel Processing

Model 1 → [Model 2A, Model 2B, Model 2C] → Merge → Output

Iterative Refinement

Model 1 → Quality Check → [Pass: Next Step | Fail: Regenerate]

Multi-Output Chain

Input → Model 1 → [Branch A → Output A, Branch B → Output B]

Workflows

Complete Product Creation

1. Generate design concept (text model)
2. Create design (image model)
3. Generate 3 variations (editing model)
4. Create mockups (mockup model)
5. Enhance quality (upscale model)
6. Create video ad (video model)
7. Add music (audio model)

Marketing Content Suite

1. Research topic (web search)
2. Write article (text model)
3. Generate hero image (image model)
4. Create social variants (editing model × 5)
5. Generate video summary (video model)
6. Add voiceover (speech model)

Dependencies

• replicate_universal: For all AI models
• execution_agent: For chain execution
• memory_agent: For chain templates

Configuration

• max_chain_length: 20
• max_parallel_models: 5
• auto_retry_failures: true
• save_intermediates: true
• chain_directory: data/model_chains/