Intelligence Nodes

Add AI reasoning, analysis, and decision-making to your workflows.

Brain

Multi-purpose AI reasoning node

The Brain node is DUTO's most versatile intelligence node. It uses large language models to analyze, reason, extract, and generate content. It features a Smart Brain architecture that auto-detects which outputs are connected and only generates what is needed downstream.

Property	Description
Inputs	ctx_in (text - Instructions), text_in (text), image_in (image), video_in (video)
Outputs	text_out (text), image_out (image), log_out (text)

Modes:

Mode	Description	Best For	Temperature
Creative	Generates creative content like stories, scripts, and marketing copy	Writing, ideas, prompts	0.9
Research	Searches and synthesizes information from web sources (Tavily)	Information extraction, web research	0.3
Hybrid	Combines research with creative generation	General tasks	0.7
Analysis	Analyzes inputs and provides structured insights	Data analysis, decisions	0.2

Configuration:

Setting	Description
Mode	Operating mode (see above)
System Prompt	Instructions for the AI
Seed	Optional seed for reproducible outputs
Model	LLM to use (Gemini, DeepSeek)

Smart Brain Architecture:

The Brain node features smart edge detection that minimizes API usage:

• Inspects the edge graph at runtime to find which output handles are connected
• Performs a single LLM call via the brain-reasoning edge function
• Extracts only the outputs that will actually be consumed downstream
• Supports multi-modal inputs: text, images, and video in a single call

Input handling:

• ctx_in (Instructions): Provides system-level instructions or context from upstream nodes
• text_in: Direct text input for processing
• image_in: Image URL for vision-based analysis
• video_in: Video URL for video analysis
• If no direct text input is provided, context input is used as fallback

Example - Prompt Enhancement:

Input: "cat on a chair"
System Prompt: "Enhance this into a detailed image prompt"
Output: "A fluffy orange tabby cat lounging elegantly
        on a vintage velvet armchair, soft natural
        lighting from a nearby window, cozy home
        interior, photorealistic style"

Example - Data Extraction:

Input: Product description text
System Prompt: "Extract product details"
Output Schema:
{
  "name": "string",
  "price": "number",
  "features": "array"
}

Use cases:

• Prompt generation and enhancement
• Web research and synthesis (Research mode with Tavily)
• Content analysis and extraction
• Decision making
• Data transformation
• Creative writing

Available Models:

Model	Strengths	Speed	Cost
Gemini 2.5 Flash	Fast, good quality	Fast	Low
Gemini 2.5 Pro	High quality	Medium	Medium
DeepSeek	Complex reasoning	Medium	Low

Credit cost: 1-5 credits depending on model and complexity

---

Analyzer

Image content analysis with AI vision

Analyzes image content using AI vision (via OpenRouter) with four analysis modes and structured output.

Property	Description
Inputs	image (image)
Outputs	description (text)

Configuration:

Setting	Description
Mode	Analysis mode (see below)

Analysis Modes:

Mode	Description
describe	General description of image content including subjects, composition, colors, lighting, and style
tags	Extracts relevant tags as a comma-separated list of keywords (subjects, style, mood, technical aspects)
recreate	Generates a detailed prompt that could recreate the image, specific about style, composition, lighting, and more
qc	Quality critique analyzing composition, framing, lighting, exposure, color, sharpness, artistic merit, and improvement areas

How it works:

1. Receives an image input (URL required)
2. Sends the image to the openrouter-chat edge function in analyze mode
3. Returns analysis text as the description output
4. Stores the result in node state for persistence
5. Propagates analysis text through the flow context for downstream nodes

Use cases:

• Automatic tagging and categorization
• Quality assessment before further processing
• Generating recreation prompts from existing images
• Content analysis for routing decisions (pair with Conditional Switch)

Credit cost: 2-6 credits

---

Story Generator

Multi-chapter AI video generation with character consistency

Creates complete multi-chapter story videos from a concept, with automatic character/subject consistency tracking across chapters.

Property	Description
Inputs	concept (text), style (text), reference (image, optional), context (text, optional)
Outputs	video (video), chapters (json), registry (json)

Configuration:

Setting	Default	Description
Chapters	6	Number of chapters to generate
Duration per Chapter	10s	Video duration per chapter (4, 6, or 8 seconds for Veo 3)
Model	google/veo3.1/video-extend	Video extension model for chapters 2+
Resolution	1080p	Video output resolution
Consistency Threshold	70	Validation score (0-100) to accept a chapter
Max Retries	3	Retry attempts per chapter if consistency fails

Generation Pipeline:

The Story Generator executes a multi-phase pipeline:

1. Phase 0 - Reference Analysis: If reference images are provided, analyzes them to build a subject registry (characters, products, vehicles, etc.)
2. Phase 1 - Story Planning: Plans the narrative with chapter breakdown using the concept and style
3. Phase 2 - Hero Image: Generates the initial image from the first chapter (uses Nano Banana Pro, or edit model if references provided)
4. Phase 3 - Registry Building: Builds a character/subject registry from the hero image for consistency tracking
5. Phase 4 - Chapter 1 Video: Converts the hero image to video using Veo 3 Fast (image-to-video)
6. Phase 5 - Chapter Extensions: Extends the video for each remaining chapter with consistency validation and retry logic

Character Consistency:

• Reference images are analyzed to build a subject registry with visual descriptions
• Each chapter prompt is enhanced with reference index blocks describing characters/subjects
• After each chapter is generated, consistency is validated against the registry
• Chapters that fail validation (below threshold) are automatically retried with enhanced prompts

Outputs:

• video: The final complete story video
• chapters: JSON array of chapter details including validation scores and attempt counts
• registry: JSON object containing the character/subject registry

Use cases:

• Automated short film creation
• Story-driven video content
• Product showcase videos with consistent branding
• Multi-scene narrative generation

Credit cost: Varies significantly based on chapter count and retries (each chapter costs credits for video generation)

---

Narrative Explorer

AI-powered video extension with narrative analysis

Analyzes the narrative context of an existing video and generates contextually appropriate extensions in any direction.

Property	Description
Inputs	video (video), context (text, optional)
Outputs	video (video - Extended Video), analysis (text - Narrative Analysis)

Configuration:

Setting	Default	Description
Direction	forward	Extension direction: forward, backward, or alternate
Model	google/veo3.1/video-extend	Video extension model
Duration	5s	Extension duration in seconds
Resolution	720p	Output resolution

Directions:

Direction	Description
Forward	Continues the story - "what happens next"
Backward	Explores the prequel - "what happened before"
Alternate	Explores a different timeline or outcome

How it works:

1. Narrative Analysis: Sends the video to the narrative-engine edge function, which analyzes the first frame to extract mood, characters, setting, and narrative context
2. Prompt Generation: The engine generates direction-appropriate extension prompts based on the analysis
3. Video Extension: Generates the video extension using the selected model and prompt
4. Polling: Polls for job completion (up to 6 minutes timeout)

Outputs:

• video: The extended video clip
• analysis: JSON containing the narrative analysis (direction, mood, narrative, protagonist, setting)

Use cases:

• Extending existing video content with narrative coherence
• Exploring "what if" scenarios with alternate timelines
• Building longer stories by chaining multiple extensions
• Creating prequels or sequels to existing footage

Credit cost: Depends on model and duration

---

Intelligence Workflow Examples

AI-Enhanced Image Generation

Use Brain to enhance prompts:

Simple Prompt --> Brain (enhance) --> Text to Image
       |               |
"sunset photo"    "Breathtaking sunset
                  over ocean, dramatic
                  clouds, golden hour..."

Content Pipeline with Analysis

Analyze and route content:

Upload --> Analyzer --> Conditional Switch --> Path A (high quality)
                              |
                        Path B (needs improvement)

Data-Driven Content

Generate content from data:

Data Feed --> Brain (transform) --> Text to Image --> Save
               |
    [Create prompts from
     product data]

Multi-Chapter Story

Generate a complete story video:

Text Prompt (concept) --> Story Generator --> Preview Media (video)
Text Prompt (style)   -->                 --> Text Preview (chapters)
Upload Media (refs)   -->                 --> Brain (process registry)

Narrative Extension Chain

Extend a video with narrative awareness:

Upload Media --> Narrative Explorer (forward) --> Narrative Explorer (forward)
                        |                              |
                  [Extends video]              [Extends again]

---

Tips for Intelligence Nodes

Brain Node

1. Clear system prompts - Be specific about what you want
2. Mode selection - Use Research for web-sourced info, Creative for generation, Analysis for structured output
3. Connect only needed outputs - Smart Brain skips unconnected outputs, saving API costs
4. Multi-modal input - Combine text with images or video for richer reasoning

Analyzer

1. Quality input - Higher quality images produce better analysis
2. Choose the right mode - Use tags for categorization, recreate for prompt generation, qc for quality assessment
3. Combine with Brain - Feed analysis output into Brain for further reasoning

Story Generator

1. Provide reference images - Character consistency improves significantly with reference images
2. Keep chapters manageable - 4-8 chapters works best; more chapters increase generation time and cost
3. Style descriptions matter - Detailed style inputs improve visual consistency across chapters

Narrative Explorer

1. Chain extensions - Connect multiple Narrative Explorers in sequence for longer stories
2. Provide context - Optional text context helps guide the narrative direction
3. Try alternate - The alternate direction can produce creative variations

Performance

Node	Typical Time	Tips
Brain	2-10s	Simpler prompts = faster
Analyzer	3-8s	Single image input only
Story Generator	5-30min	Depends on chapter count
Narrative Explorer	1-6min	Depends on video model