Image Generation Node

The Image Generation component enables workflows to generate images dynamically using generative AI models. It converts natural language prompts into visual outputs and optionally stores generated images in cloud storage providers such as AWS, Azure, or GCP.

---

Component Introduction

Field	Value
Component Type	image_generation
Input	Natural language text prompt
Output	Generated image — URL
Supported Models	Gemini image models
Storage Options	Local response or Cloud (AWS / Azure / GCP)

---

Core JSON Structure

{
  "name": "Test Image Generation",
  "type": "image_generation",
  "description": "Test image gen component",
  "output_type": "json",
  "inputs": {
    "user_query": "a girl driving a red car",
    "model": "gemini_3_image",
    "variations": "1",
    "size": "1024x1024",
    "cloud_provider": "aws",
    "use_cloud": false
  }
}

---

How to Initialize

Step 1: Add the Node

Drag the Image Generation component from the Tools section onto the workflow canvas.

Step 2: Enter a Prompt

Set the user_query field to describe the image you want generated.

"a futuristic city at sunset"

Step 3: Select a Model

Choose an image generation model from the dropdown — for example, gemini_2_5_image or gemini_3_image.

Step 4: Set Image Properties

Configure resolution and output count:

• Size — output resolution, e.g. 1024x1024
• Variations — number of images to generate per run

Step 5: Configure Cloud Storage (Optional)

Enable use_cloud: true to persist generated images externally.

• Select a provider: aws, azure, or gcp
• Provide the required credentials and bucket/storage details

Step 6: Connect Ports

• Input port — connect to a prompt source (user input node or upstream LLM node)
• Output port — connect to downstream processing (API call, storage, or UI display)

---

Input Parameters

Parameter	Description
user_query	Natural language prompt describing the image
model	Image generation model to use
size	Output resolution (e.g. 1024x1024)
variations	Number of images to generate
use_cloud	Whether to upload output to cloud storage
cloud_provider	aws, azure, or gcp
azure_* / gcp_* fields	Provider-specific credentials and configuration

---

Cloud Storage Behaviour

use_cloud	Result
false	Image is stored in Kompass cloud storage(AWS)
true	Image uploaded to your configured cloud storage; output includes URL

Output typically includes the image URL or path, model used, resolution, and generation metadata.

---

Prompt Engineering Tips

The quality of your output depends directly on the quality of your prompt.

	Example
Vague	"A girl playing tennis"
Descriptive	"A young girl playing table tennis on a sunny street, cinematic lighting, high detail"

Include subject, environment, lighting, style, and detail level for best results.

---

Do's and Don'ts

Do's

• Write clear, descriptive prompts — include subject, setting, and style
• Use supported resolutions — stick to known sizes like 1024x1024
• Enable cloud storage for production workflows to persist outputs
• Test with variations — multiple outputs improve creative coverage
• Store credentials securely — use the Vault, never hardcode secrets

Don'ts

• Avoid vague prompts — generic descriptions produce generic results
• Never expose private keys in logs, UI, or workflow configs
• Avoid unsupported sizes — incorrect dimensions will fail execution
• Don't set high variation counts unnecessarily — increases cost and latency
• Don't skip the Playground — always validate output before deploying

---

Common Constraints

• Image models may only support fixed resolutions — check model documentation before setting custom sizes
• Latency increases with higher resolution and more variations
• Cloud upload requires valid credentials and appropriate storage permissions

---

Power Pattern: LLM → Image Generation

Use an LLM node to dynamically generate a detailed prompt, then pipe its output directly into the user_query field of the Image Generation node. This enables fully automated, context-aware image creation.