Gilgen Textbox Workflow Guide for Promptus

What is GLIGEN?

GLIGEN (Grounded Language-to-Image Generation) is an advanced AI model that extends traditional text-to-image generation by adding precise spatial control. Instead of just describing what you want in your image, GLIGEN lets you specify exactly where objects should appear using bounding boxes.

Key Benefits:

• Precise Object Placement: Control exactly where elements appear in your image
• Multiple Object Control: Position several different objects simultaneously
• Enhanced Creative Control: Move beyond basic text prompts to detailed spatial composition

Who Created GLIGEN?

GLIGEN was developed by researchers and engineers from:

• University of Wisconsin-Madison
• Columbia University
• Microsoft

The model builds upon existing diffusion models like Stable Diffusion while adding groundbreaking spatial conditioning capabilities.

How GLIGEN Works

Core Concept

GLIGEN works by combining two types of input:

1. Text Prompts - Describe what you want to generate
2. Bounding Boxes - Define where specific objects should be placed

The model uses a frozen CLIP text encoder and injects spatial information through special trainable layers, preserving the vast knowledge of pre-trained models while adding precise control.

Coordinate System

• Origin Point: Top-left corner (0,0)
• X-axis: Increases moving right
• Y-axis: Increases moving down
• Units: Pixel coordinates

Using GLIGEN in Promptus CosyFlows

What is CosyFlows?

CosyFlows is Promptus's user-friendly interface that makes ComfyUI accessible to non-technical users through:

• Drag-and-drop workflow building
• Visual feedback and previews
• Pre-built templates
• Zero-code approach

Step-by-Step Workflow

1. Set Up Your Base Workflow

• Start with a standard text-to-image workflow in CosyFlows
• Include your basic components: model loader, text prompt, sampler, VAE decoder

2. Add the (cosy) GLIGEN Textbox Node

• Drag the GLIGEN Textbox Apply node into your workflow
• This is Promptus's user-friendly version of the standard GLIGEN node

3. Configure Your Main Prompt

• Write your complete image description as you normally would
• Example: "A serene beach scene with a red umbrella, a golden retriever, and a sailing boat"

4. Set Up Spatial Controls

For each object you want to position precisely:

Input Fields:

• Text: The specific object name (must match something in your main prompt)
• Width: How wide the bounding box should be (in pixels)
• Height: How tall the bounding box should be (in pixels)
• X Position: Horizontal position from left edge
• Y Position: Vertical position from top edge

Example Configuration:

Object 1: "red umbrella"
- Width: 200, Height: 300
- X: 100, Y: 50

Object 2: "golden retriever"
- Width: 250, Height: 200
- X: 400, Y: 300

Object 3: "sailing boat"
- Width: 300, Height: 150
- X: 600, Y: 100


5. Connect the Nodes

• Connect your CLIP model to the GLIGEN node
• Connect the GLIGEN model to the workflow
• Link the conditioning output to your sampler

6. Generate and Iterate

• Run the workflow to see your spatially-controlled generation
• Adjust bounding boxes as needed for better composition
• Use the "Remix" feature to generate variations

Best Practices

Prompt Writing

• Be Specific: Use clear, descriptive object names
• Match References: Ensure GLIGEN text exactly matches objects in your main prompt
• Avoid Overlap: Try not to place bounding boxes directly on top of each other

Bounding Box Sizing

• Proportional Sizing: Make boxes appropriate for the object size
• Leave Breathing Room: Don't make boxes too tight around objects
• Consider Context: Account for how objects interact with the background

Workflow Optimization

• Start Simple: Begin with 1-2 objects, then add more
• Use Templates: Leverage Promptus's pre-built GLIGEN workflows
• Save Successful Setups: Create reusable workflows for common compositions

Troubleshooting Tips

Objects Not Appearing in the Right Place:

• Check that your text exactly matches the main prompt
• Verify coordinate values are within image boundaries
• Try larger bounding boxes

Poor Quality Results:

• Ensure your base prompt is well-written
• Check that the GLIGEN model is properly loaded
• Adjust sampling settings if needed

Objects Bleeding Outside Boxes:

• Increase the strength of spatial conditioning
• Use more specific object descriptions
• Try adjusting the guidance scale

Advanced Techniques

Multiple GLIGEN Nodes

• Chain multiple GLIGEN Textbox nodes for complex scenes
• Each node can control different objects independently
• Build up compositions layer by layer

Combining with Other Nodes

• Use with ControlNet for additional guidance
• Combine with inpainting for refined results
• Integrate with style transfer nodes

Batch Generation

• Use Promptus's batch features to generate multiple variations
• Test different spatial arrangements quickly
• Create lookbooks or style guides efficiently

Getting Started on Promptus

1. Sign up at promptus.ai
2. Browse the Gallery for GLIGEN workflow templates
3. Fork a Template or create a new CosyFlow
4. Experiment with the (cosy) GLIGEN Textbox node
5. Share and Remix your successful workflows with the community

Conclusion

GLIGEN in Promptus CosyFlows democratizes advanced spatial control in AI image generation. Whether you're a designer creating precise layouts, an artist exploring composition, or a marketer needing consistent product placement, GLIGEN provides the tools to transform your creative vision into reality with unprecedented control and ease.