Referenced Images in Prompts

Sometimes your prompt needs to use the same image every time it runs—for example:

• A brand style guide.
• A reference UI mockup.
• A logo sheet or photo you want the AI to analyze or keep in mind.

Instead of having to upload that image on every run, you can attach it once as a referenced image and wire it into your prompt logic.

When to use a referenced image

Use a referenced image when:

• The image is part of the prompt itself, not unique per user.
• Every run of the prompt should consider or analyze the same visual reference.
• You want a consistent visual baseline (e.g. "match this style" or "evaluate against this design").

If each user needs to upload their own image for every run, use a regular image input instead.

How referenced images differ from user image inputs

• Referenced image: Attached once by the author, reused silently on every run. The user does not need to upload it.
• User image input: The runner uploads or selects an image each time; it changes from run to run.

Both ultimately feed an image into the underlying model, but referenced images behave more like fixed context than user‑supplied data.

Attaching a referenced image to your prompt

To create a prompt that always uses the same image:

1. Prepare your image asset
   • Use a clear, high‑quality image (PNG or JPG is usually best).
   • Make sure it's cropped or framed so the important content is obvious.
2. Attach the image in the prompt editor
   • In the prompt editor, use the image upload/attach control for a referenced image (not a user input).
   • Give it a meaningful internal name (for example, brand_style_guide or dashboard_mockup).
3. Save the prompt
   • The builder will store the image and include it whenever the prompt runs.

From the runner's perspective, they just see and fill out the text inputs; the image is automatically included behind the scenes.

Referencing the image in your prompt text

Once your referenced image is attached, update your hidden logic so the model knows how to use it.

You are a visual design assistant.

You are given a reference image that shows our typography, colors, and spacing.
Use that image as the source of truth when evaluating or suggesting designs.

Based on the reference image and the description below, provide feedback or recommendations:
{{ user_request }}

Tips:

• Clearly tell the model what the image contains and how it should influence the answer.
• Use consistent naming so your mental model of the prompt matches what the builder is sending.

Common use cases

Referenced images work well for:

• Branding: "Compare this new design concept to our brand style guide image."
• Product UX: "Review this new screen concept relative to the reference dashboard layout."
• Quality control: "Check this screenshot against the reference and list differences or issues."

Anywhere you want the model to "know" a fixed visual reference, a referenced image keeps you from re‑uploading it every time.

Combining referenced images with user inputs

You can still mix in regular inputs alongside a referenced image, for example:

• A text field for {{ user_request }} describing what feedback they want.
• A dropdown or text input for {{ audience }} or {{ use_case }}.
• An image input for {{ item }} to redesigned in the referenced image's style.

The pattern looks like:

1. The referenced image provides the shared context.
2. The placeholder inputs provide run‑specific details.
3. The model uses both to generate a consistent, contextual answer.

Design your prompt so users only have to think about what they want from the AI—while the image is quietly doing its job in the background.