Workflow Patterns

To build any example: – press Space to add nodes – drag connections – press Ctrl/⌘+Enter to run – add Preview nodes to inspect intermediate results

Pattern 1: Simple Pipeline

Use Case: Transform input → process → output

Example: Image Enhancement

graph TD output["Output"] image_input["ImageInput"] sharpen["UnsharpMask"] auto_contrast["AutoContrast"] image_input --> sharpen sharpen --> auto_contrast auto_contrast --> output

When to Use:

  • Simple data transformations
  • Single input, single output
  • No conditional logic needed

Pattern 2: Agent-Driven Generation

Use Case: LLM generates content based on input

Example: Image to Story

graph TD image_input["Image"] agent_story["Agent (Story Generator)"] preview_audio["Preview (Audio)"] text_to_speech["TextToSpeech"] image_input --> agent_story agent_story --> text_to_speech text_to_speech --> preview_audio

When to Use:

  • Creative generation tasks
  • Multimodal transformations (image→text→audio)
  • Need semantic understanding

Key Nodes:

  • Agent: General-purpose LLM agent with streaming
  • Summarizer: Specialized for text summarization
  • ListGenerator: Streams list of items

Pattern 3: Streaming with Multiple Previews

Use Case: Show intermediate results during generation

Example: Movie Poster Generator

graph TD strategy_llm["Agent (Strategy)"] strategy_template_prompt["String (Strategy Template)"] strategy_preview["Preview (Strategy)"] strategy_prompt_formatter["FormatText (Strategy)"] movie_title_input["StringInput (Title)"] genre_input["StringInput (Genre)"] audience_input["StringInput (Audience)"] image_preview["Preview (Image)"] prompt_list_generator["ListGenerator"] designer_instructions_prompt["String (Designer Instructions)"] preview_prompts["Preview (Prompts)"] text_to_image["TextToImage"] strategy_llm --> strategy_preview strategy_template_prompt --> strategy_prompt_formatter strategy_prompt_formatter --> strategy_llm strategy_llm --> prompt_list_generator designer_instructions_prompt --> prompt_list_generator audience_input --> strategy_prompt_formatter movie_title_input --> strategy_prompt_formatter genre_input --> strategy_prompt_formatter prompt_list_generator --> preview_prompts prompt_list_generator --> text_to_image text_to_image --> image_preview

When to Use:

  • Complex multi-stage generation
  • User needs to see progress
  • Agent planning + execution workflow

Key Concepts:

  • Strategy Phase: Agent plans approach
  • Preview Nodes: Show intermediate results
  • ListGenerator: Streams generated prompts
  • Image Generation: Final output

Pattern 4: RAG (Retrieval-Augmented Generation)

Use Case: Answer questions using documents as context

Example: Chat with Docs

graph TD chat_input["StringInput"] output["Output (Answer)"] format_text["FormatText"] hybrid_search["HybridSearch"] agent["Agent"] chat_input --> format_text chat_input --> hybrid_search hybrid_search --> format_text format_text --> agent agent --> output

When to Use:

  • Question-answering over documents
  • Need factual accuracy from specific sources
  • Reduce LLM hallucinations

Key Components:

  1. Search: Query vector database for relevant documents
  2. Format: Inject retrieved context into prompt
  3. Generate: Stream LLM response with context

Pattern 5: Database Persistence

Use Case: Store generated data for later retrieval

Example: AI Flashcard Generator with SQLite

graph TD topic_input["StringInput (Topic)"] create_table["CreateTable"] format_prompt["FormatText"] generate_flashcards["DataGenerator"] insert_flashcard["Insert"] query_all["Query"] display_result["Preview"] topic_input --> format_prompt format_prompt --> generate_flashcards generate_flashcards --> insert_flashcard create_table --> query_all create_table --> insert_flashcard query_all --> display_result

When to Use:

  • Need persistent storage
  • Building apps with memory
  • Agent workflows that need to recall past interactions

Key Nodes:

  • CreateTable: Initialize database schema
  • Insert: Add records
  • Query: Retrieve records
  • Update: Modify records
  • Delete: Remove records

Database Flow:

  1. Create table structure
  2. Generate data with agent
  3. Insert into database
  4. Query and display results

Pattern 6: Email & Web Integration

Use Case: Process emails or web content

Example: Summarize Newsletters

graph TD gmail_search["GmailSearch"] email_fields["Template"] summarizer_streaming["Summarizer"] preview_summary["Preview (Summary)"] preview_body["Preview (Body)"] gmail_search --> email_fields email_fields --> summarizer_streaming summarizer_streaming --> preview_summary email_fields --> preview_body

When to Use:

  • Automate email processing
  • Monitor RSS feeds
  • Extract web content

Key Nodes:

  • GmailSearch (lib.mail.GmailSearch): Search Gmail with queries
  • Template: Format email fields into text (lib.mail also provides AddLabel, MoveToArchive, SendEmail)
  • FetchRSSFeed: Get RSS feed entries
  • GetRequest: Fetch web content

Pattern 7: Multi-Modal Workflows

Use Case: Convert between different media types

Example: Audio to Image

graph TD stable_diffusion["StableDiffusion"] whisper["Whisper"] audio_input["AudioInput"] output["Output"] whisper --> stable_diffusion audio_input --> whisper stable_diffusion --> output

When to Use:

  • Converting between media types
  • Creating rich multimedia experiences
  • Accessibility applications

Common Chains:

  • Audio → Text → Image
  • Image → Text → Audio
  • Video → Audio → Text → Summary

Pattern 8: Advanced Image Processing

Use Case: AI-powered image transformations

Example: Style Transfer

graph TD sd_img2img["StableDiffusionV3MediumImageToImage"] image_input_1["ImageInput"] image_input_2["ImageInput"] output["Output"] image_to_text["ImageToText"] fit_1["Fit"] fit_2["Fit"] sd_img2img --> output image_to_text --> sd_img2img image_input_2 --> fit_1 image_input_1 --> fit_2 fit_2 --> image_to_text image_input_2 --> sd_img2img fit_2 --> sd_img2img

When to Use:

  • Style transfer between images
  • Controlled image generation
  • Preserving structure while changing style

Key Techniques:

  • Img2Img: Transform while maintaining composition (StableDiffusionV3MediumImageToImage, or StableDiffusion / StableDiffusionXL for text-to-image)
  • Image-to-Text: Generate descriptions (ImageToText)
  • Canny: Edge detection preprocessing

Pattern 9: Text-to-Video Generation

Use Case: Generate videos from text descriptions

Example: Cinematic Video from Prompt

graph TD string_input["StringInput (Prompt)"] output["Output"] kling_text_to_video["KlingVideoV16ProTextToVideo"] string_input --> kling_text_to_video kling_text_to_video --> output

When to Use:

  • Create videos from text descriptions
  • Generate concept videos and storyboards
  • Produce cinematic content from prompts
  • Rapid video prototyping

Key Nodes:

  • KlingVideoV16ProTextToVideo: High-quality text-to-video (Kling 1.6 Pro)
  • MinimaxHailuo23ProTextToVideo: Professional quality (Hailuo 2.3)
  • Sora2TextToVideo: OpenAI Sora 2 model
  • WanProTextToVideo: Alibaba Wan

Configuration Tips:

  • Duration: 5-10 seconds for most models
  • Resolution: 768P for faster generation, 1080P for quality
  • Aspect ratios: 16:9 (landscape), 9:16 (portrait), 1:1 (square)

Pattern 10: Image-to-Video Generation

Use Case: Animate images into videos

Example: Bring Images to Life

graph TD image_input["ImageInput"] string_input["StringInput (Motion Guide)"] output["Output"] kling_image_to_video["KlingVideoV16StandardImageToVideo"] image_input --> kling_image_to_video string_input --> kling_image_to_video kling_image_to_video --> output

When to Use:

  • Animate static images
  • Create motion from photographs
  • Multi-image video generation
  • Product showcases from images

Key Nodes:

  • KlingVideoV16StandardImageToVideo: Kling 1.6 Standard image-to-video
  • MinimaxHailuo02ProImageToVideo: High-quality animation (Hailuo 02 Pro)
  • SeeDanceV15ProImageToVideo: ByteDance SeeDance 1.5 Pro
  • WanV225bImageToVideo: Alibaba Wan 2.2

Advanced Pattern: Multi-Image Animation

graph TD image1["ImageInput (Frame 1)"] image2["ImageInput (Frame 2)"] image3["ImageInput (Frame 3)"] motion_prompt["StringInput (Motion)"] output["Output"] kling_i2v["KlingVideoV16StandardImageToVideo"] kling_i2v --> output image1 --> kling_i2v image2 --> kling_i2v image3 --> kling_i2v motion_prompt --> kling_i2v

Pattern 11: Talking Avatar Generation

Use Case: Create lip-synced avatar videos

Example: Virtual Presenter

graph TD image_input["ImageInput (Face Photo)"] audio_input["AudioInput (Speech)"] output["Output"] kling_avatar["KlingVideoAiAvatarV2Pro"] image_input --> kling_avatar audio_input --> kling_avatar kling_avatar --> output

When to Use:

  • Create virtual presenters
  • Generate lip-synced avatar videos
  • Educational content with AI speakers
  • Virtual influencers and spokespersons

Key Nodes:

  • KlingVideoAiAvatarV2Pro: Pro-quality avatar generation
  • KlingVideoAiAvatarV2Standard: Standard mode for faster generation
  • Infinitalk: Audio-driven video generation

Workflow: Audio + Image → Talking Avatar

  1. Photo Input: Front-facing portrait image
  2. Audio Track: Speech recording or TTS output
  3. Optional Prompt: Guide emotions and expressions
  4. Mode Selection: Standard (faster) or Pro (higher quality)

Pattern 12: Image Enhancement & Upscaling

Use Case: Improve image quality and resolution

Example: HD Image Upscaling

graph TD image_input["ImageInput (Low Res)"] output["Output (High Res)"] topaz_upscale["TopazUpscaleImage"] image_input --> topaz_upscale topaz_upscale --> output

When to Use:

  • Upscale low-resolution images
  • Remove noise and artifacts
  • Enhance image quality
  • Prepare images for high-resolution displays

Key Nodes:

  • TopazUpscaleImage: AI-powered upscaling to higher resolutions
  • Denoise option: Reduces artifacts during upscaling

Configuration:

  • Higher target resolutions for crisp output
  • Denoise: Enable for noisy input images
  • Best for: Enhancing old photos, smartphone images, web graphics

Pattern 13: Storyboard to Video

Use Case: Convert image sequences to coherent videos

Example: Visual Story Generation

graph TD story_prompt["StringInput (Story)"] image1["ImageInput (Scene 1)"] image2["ImageInput (Scene 2)"] image3["ImageInput (Scene 3)"] output["Output"] sora_storyboard["Sora2ImageToVideoPro"] sora_storyboard --> output story_prompt --> sora_storyboard image1 --> sora_storyboard image2 --> sora_storyboard image3 --> sora_storyboard

When to Use:

  • Create narrative videos from storyboards
  • Combine multiple scenes into one video
  • Professional video pre-visualization
  • Animated story creation

Key Nodes:

  • Sora2ImageToVideoPro: OpenAI Sora 2 Pro keyframe-driven generation
  • Supports: 1-3 keyframe images
  • Output: Smooth transitions between scenes

Workflow:

  1. Story Prompt: Describe the narrative arc
  2. Keyframes: Provide 1-3 scene images
  3. Generation: Sora creates smooth transitions
  4. Duration: 1-60 frames configurable
Edit this page on GitHub