Generating AI images with precise control over character posture has historically been difficult. However, by combining the structural guidance of OpenPose with the advanced image synthesis capabilities of Google Nano Banana, you can achieve unparalleled control. This guide explains what these technologies are and how to use them together.
Quick Start: The Magic Prompt
If you already have your images ready, use this prompt with Nano Banana (in Gemini Advanced or AI Studio):
I have uploaded two images. You must identify which is which based on their visual content:
1. **Identify the Pose Reference:** Look for the image that resembles a stick figure, skeleton, or color-coded OpenPose map. Use this image strictly to control the body position, limb angles, and gesture.
2. **Identify the Character Reference:** Look for the image showing a fully rendered character/person. Use this image as the source for the identity, facial features, hair, clothing, and overall style.
**Task:**
Generate a new, high-fidelity image of the character from the **Character Reference**, but strictly performing the action shown in the **Pose Reference**.
- Maintain the character's exact appearance (eye color, hair color, outfit details).
- Maintain the pose's exact structure.
What are Nano Banana and OpenPose?
To master this workflow, it helps to understand the tools you are working with.
Google Nano Banana
Nano Banana (often referred to as Nano Banana Pro or Gemini 3 Pro Image) is Google's latest state-of-the-art image generation model. It excels at understanding complex instructions, maintaining character consistency, and rendering high-fidelity details. Unlike older models that might struggle to blend two distinct visual inputs, Nano Banana has a high "reasoning" capability that allows it to distinguish between structural inputs and style inputs.
OpenPose
OpenPose is a popular library for real-time multi-person keypoint detection. In the context of AI art, an "OpenPose image" is a simplified skeleton map—typically a black background with colorful lines representing the limbs and torso.
- Why use it? Instead of describing a pose with words (e.g., "standing with one leg up and arm waving"), which can be ambiguous, an OpenPose image provides exact geometric coordinates for the model to follow.
The Workflow
Prepare Your Character Reference: Choose a clear, high-quality image of the character you want to generate. This defines who is in the picture.
Prepare Your Pose Reference: Obtain an OpenPose skeleton image. This defines what the character is doing. We recommend using the free online tool OpenPoseAI.com.
How to use OpenPoseAI.com:
- Visit the site: Go to openposeai.com.
- Detect from Image: To extract a pose from an existing image, click "File" in the menu, then select "Detect from image" and upload your desired image. The tool will then generate an OpenPose skeleton from it.
- Adjust the Skeleton: (Optional) Click and drag the colorful joint dots to fine-tune the pose exactly how you want. You can control the angle of the head, arms, legs, and even feet.
- Save the Pose: Locate the "overview triangle button" (typically in the bottom right corner). Click it to "snapshot" the current pose. Then, click on the generated snapshot to download the pose reference image (a black background with colored lines).
Feed Both to Nano Banana: Upload both images to the Gemini App or Google AI Studio.
Apply the Prompt: Paste the prompt provided in the "Quick Start" section above.
Why This Works
Nano Banana's multimodal capabilities allow it to "see" the skeleton image not as art, but as a diagram. By explicitly instructing the model to treat the OpenPose image as a constraint for "body position" and the character image as the source for "identity," you effectively decouple structure from style. This results in a generated image that looks exactly like your character, doing exactly what you wanted them to do.
