How to Prompt and Annotate Multiple Images with AI | RunDiffusion

RunDiffusion offers cloud-based GPU workspaces and pre-configured AI tools designed to support professional creative workflows.

Multi-image support within RunDiffusion enables advanced editing and compositing workflows. When working with more than one image, it is essential to clearly define which image will be modified and which image or images provide reference content such as objects, subjects, or styles.

This guide explains how to structure multi-image prompts in the RunDiffusion platform, to explicitly target the correct image and use annotations effectively. These practices reduce ambiguity and produce more consistent, predictable results.

1. Reference and define the role of each image explicitly

When multiple images are uploaded, the AI does not automatically understand user intent. It interprets instructions based on the prompt, selected model, and image content. If instructions are unclear, the system may insert objects into the wrong image, merge scenes unintentionally, replace content instead of adding it, or apply styles to the incorrect image.

Explicit image targeting ensures that edits are applied in the correct context.

2. Apply the best practices for multi-image editing

• Explicitly reference images, annotations, or labeled areas in the prompt
• Keep each generation focused on a small number of changes
• Avoid combining multiple edits in a single prompt
• Use multiple iterations instead of one complex instruction
• Pair clear image targeting with short, direct prompts

Clear structure improves consistency and reduces unintended edits.

3. Always specify the target image

Keep in mind that every multi-image prompt must clearly answer two questions. First, which image is being modified? And second, which image provides the reference object, subject, or style? Failure to define these roles introduces ambiguity and reduces accuracy.

• Example: Apple and Bowl

A user uploads two images: one showing an apple and another showing a bowl. Their goal is to place the apple inside the bowl.

An unclear prompt, such as “Combine the images,” may produce unpredictable results. The AI could place the bowl on top of the apple, merge both scenes incorrectly, or apply the edit to the wrong image.

• Corrected prompt targeting the apple image: "Add the bowl to the image with the apple and place the apple inside the bowl."
• Corrected prompt targeting the bowl image: "Add the apple to the image with the bowl and place the apple inside the bowl."

These prompts remove ambiguity by explicitly defining the target image.

4. Enhance the prompt structure

Use the following formula for improved accuracy across tools and models:

• "Action + object or reference + source image description + target image description + placement details"
• I.e., "Add the chair from the modern living room reference image into the dining room scene, placed next to the wooden table."

Avoid generic identifiers such as “Image 1” unless clearly labeled within the interface. Instead, describe images by content. This structure aligns intent with execution and improves predictability.

5. Use annotations for precision

Annotations provide visual reinforcement for written prompts and are especially useful in complex scenes. RunDiffusion includes native annotation tools within the platform.

Common annotation methods include:

• Circling: Identify objects or areas to include or modify
• Crossing out: Mark elements to remove or ignore
• Numbered labels: Support structured, multi-step edits
• Arrows: Indicate placement or specific regions for modification

Annotations should complement, not replace, written instructions. Always describe how annotations should be interpreted. For example:

• Follow and remove the annotations marked on the image. Replace the areas indicated by arrows with black basalt. Do not change any other part of the image.

Reference	Output

• Follow and remove the annotations marked on the image; add accent lighting as shown in the green areas from my reference image to the apartment room, where it is shown in red.

RunDiffusion platform

Reference 1	Reference 2

Final Output

6. Avoid these common mistakes

• Referring to images as “this” or “that”
• Relying on the upload order alone
• Combining multiple actions without defining targets
• Forgetting to specify which image is modified
• If a prompt becomes complex, divide it into multiple steps

Multi-image prompting relies on clarity rather than complexity. By explicitly defining the target image, the role of each reference image, and the scope of each modification, users gain greater control over AI-driven workflows. Clear prompts enable faster iteration, fewer errors, and more consistent results across RunDiffusion.

Explore RunDiffusion's product catalog or visit their website for more information.