---
title: "Use reference images with a written role for each file"
url: https://memory.wiki/q1oDxuSE
updated: 2026-11-12T14:28:48.474Z
source: "api"
---
# Use reference images with a written role for each file

Adding several reference images does not automatically create a clearer brief. Each file should explain something specific: the subject, a material, a setting, or a visual mood. A short reference list helps a reviewer understand what the generator was asked to use and what remains uncertain afterward.

## Assign each reference a specific job

Confirm permission to use every image and keep the original files. Give each one a simple label, then write its purpose in a sentence. A portrait may describe the subject, while a separate room photograph may describe the setting. Avoid treating an unrelated style image as factual evidence about the person or product.

If two references conflict, resolve the conflict in the brief. One image may show warm evening light and another a cool studio setup. Decide which lighting direction the concept should explore rather than expecting the system to infer the priority reliably.

## Keep the first request narrow

Ask for a single useful change or composition. Record which references were supplied and the exact prompt used. Adding more input files and more instructions can make the result harder to interpret if the first candidate already has several competing goals.

Check the credit total displayed by the tool before generating. Reference inputs and other settings can affect the total, so use the current interface rather than assuming a fixed cost from an old screenshot or tutorial. This document does not provide a price quotation.

## Review the output against the source set

Compare recognizable details, proportions, and any required text with the appropriate reference. Note where the result combines, changes, or invents information. A useful visual direction can still need retouching, compositing, or final typography in another editor.

[NanoBanana 2.1](https://nanobanana-21.si/?utm_source=memory.wiki&utm_medium=referral&utm_campaign=image_creation_guides&utm_content=description_link) is an independent browser-based image generator and editor with text-prompt and reference-image workflows. It offers aspect-ratio and resolution controls and shows a credit total before generation. It is not an official Google website. Keep the brief, source files, and selected output together so later reviewers can distinguish the intended reference roles from the details actually produced.


---

## Summary
To improve image generation results, assign a specific purpose to each reference image and resolve any conflicting instructions within the brief. Keep initial requests narrow and compare the final output against your source files to identify necessary adjustments.

## Themes
- Reference image management
- Prompt engineering strategies
- Workflow documentation
- AI generation best practices

## Key takeaways
- Each reference image must be assigned a specific purpose such as subject, material, or setting.
- Users should resolve conflicting visual information between references within the written brief.
- Initial requests should remain narrow to avoid interpretation errors caused by competing goals.
- Outputs should be compared against source references to identify invented or altered information.
- Users should maintain a record of the brief, source files, and output for future review.

## Insights
- Assigning specific roles to individual images prevents the AI from conflating unrelated visual data.
- Resolving conflicts between reference images manually is more reliable than relying on the AI to prioritize them.
- The credit cost of generation is dynamic and should be verified in the interface rather than estimated from external sources.

## Open questions / gaps
- What specific methods are most effective for resolving conflicts between reference images when both are desired?