Back to Templates

Generate Images from Text with IBM Granite Vision 3.3 2B AI Model

Created by

Created by: Yaron Been || yaron-nofluff

Yaron Been

Last update

Last update 15 days ago

Share


Generate Images from Text with IBM Granite Vision 3.3 2B AI Model

🌍 Overview

This workflow uses the ibm-granite/granite-vision-3.3-2b model (hosted on Replicate) to generate AI images. It starts manually, sends a request to the Replicate API, waits for the result, and finally outputs the generated image link.

Think of it as your AI art assistant — you click once, and it handles the full request/response cycle for image generation.


🟢 Section 1: Trigger & API Setup

🔗 Nodes:

  • Manual Trigger → Starts when you click Execute.
  • Set API Key → Stores your Replicate API Key safely in the workflow.

💡 Beginner takeaway:
This section is like turning the key in the ignition. You start the workflow, and it loads your credentials so you can talk to Replicate’s API.

📈 Advantage:
Keeps your API key stored inside the workflow instead of hard-coding it everywhere.


🟦 Section 2: Create Prediction

🔗 Nodes:

  • HTTP Request (Create Prediction) → Sends a request to Replicate with the chosen model (granite-vision-3.3-2b) and input parameters (seed, temperature, max_tokens, etc.).

💡 Beginner takeaway:
This is where the workflow actually asks the AI model to generate an image.

📈 Advantage:
You can tweak parameters like creativity (temperature) or randomness (seed) to control results.


🟣 Section 3: Polling & Status Check

🔗 Nodes:

  • Extract Prediction ID (Code) → Saves the unique job ID.

  • Wait (2s) → Pauses before checking status.

  • Check Prediction Status (HTTP Request) → Calls Replicate to see if the image is ready.

  • If Condition (Check If Complete)

    • ✅ If status = succeeded → move to result
    • 🔄 Else → go back to Wait and check again

💡 Beginner takeaway:
Since image generation takes a few seconds, this section keeps asking the AI “are you done yet?” until the image is ready.

📈 Advantage:
No need to guess — the workflow waits automatically and retries until success.


🔵 Section 4: Process Result

🔗 Nodes:

  • Process Result (Code) → Extracts the final data:

    • ✅ Status
    • ✅ Output image URL
    • ✅ Metrics (time taken, etc.)
    • ✅ Model info

💡 Beginner takeaway:
This section collects the finished image link and prepares it neatly for you.

📈 Advantage:
You get structured output that you can save, display, or use in another workflow (like auto-sending images to Slack or saving to Google Drive).


📊 Final Overview Table

Section Nodes Purpose Benefit
🟢 Trigger & Setup Manual Trigger, Set API Key Start + load credentials Secure API key management
🟦 Create Prediction HTTP Request Ask AI to generate image Control creativity & output
🟣 Polling Extract ID, Wait, Check Status, If Repeatedly check job status Auto-wait until done
🔵 Process Result Process Result Extract image + details Get clean output for reuse

🚀 Why This Workflow is Useful

  • Automates full API cycle → From request to final image URL
  • Handles delays automatically → Keeps checking until your image is ready
  • Customizable parameters → Adjust creativity, randomness, and token limits
  • Reusable → Connect it to email, Slack, Notion, or storage for instant sharing
  • Beginner-friendly → Just plug in your API key and hit Execute