Glif Workflow Builder

Design AI workflows using Glif's block-based system.

Workflow Structure

Glifs are JSON graphs with nodes (blocks) connected via variable references.

{
  "nodes": [
    {"name": "input1", "type": "TextInputBlock", "params": {...}},
    {"name": "generator", "type": "ImageBlock", "params": {"prompt": "{input1}"}}
  ]
}

Block Types

Input Blocks

• TextInputBlock - Text input from user
• ImageInputBlock - Image upload
• MultipickBlock - Multiple choice selector

AI Generation Blocks

• TextBlock - LLM text generation (GPT, Claude, Llama)
• ImageBlock - Image generation (SDXL, Flux, DALL-E)
• ImageToTextBlock - Image captioning/analysis
• VideoBlock - Video generation

Utility Blocks

• CombineTextBlock - Concatenate text values
• JSONExtractorBlock - Parse JSON responses
• PublishedGlifBlock - Nest other glifs
• WebFetchBlock - HTTP requests

Output Blocks

• HTMLBlock - Custom HTML rendering
• CanvasBlock - Image composition
• AudioBlock - Audio output

Variable References

Reference other blocks using {blockName}:

{
  "name": "enhancer",
  "type": "TextBlock",
  "params": {
    "prompt": "Enhance this prompt: {userInput}",
    "model": "gpt-4"
  }
}

Design Patterns

Basic Image Generator

TextInput → ImageBlock → Output

Prompt Enhancement

TextInput → TextBlock (enhance) → ImageBlock → Output

Multi-Image Composite

TextInput → ImageBlock (bg) ─┐
TextInput → ImageBlock (fg) ─┼→ CanvasBlock → Output

Iterative Refinement

TextInput → ImageBlock → ImageToTextBlock → TextBlock → ImageBlock → Output

Best Practices

1. Keep inputs minimal - 2-3 max for usability
2. Name blocks descriptively - stylePrompt not text1
3. Use alphanumeric + underscore only for names
4. Chain text blocks for complex prompts
5. Test incrementally - add one block at a time

Schema Reference

Full JSON schema: https://glif.app/api/graphJsonSchema

Common Models

Text: gpt-4, gpt-4o, claude-3, llama-3 Image: flux-pro, sdxl, dall-e-3, stable-diffusion-3 Video: runway-gen3, kling