Practical Adobe AI – Lesson 03: A Prompt Framework for Design Work

This practical lesson focuses on Turn a vague brief into subject, setting, composition, lighting, style, and constraint clauses. The output is Ten versioned prompts with evaluation criteria.. Treat AI output as editable material. Keep the source, settings, model name, review notes, and final export so another person can reproduce the decision path.

Feature availability, labels, credit cost, and interface placement may vary by plan, platform, region, and release. Last documentation check: 2026-09-29.

1. Learning outcomes

• Turn a vague brief into subject, setting, composition, lighting, style, and constraint clauses.
• Ten versioned prompts with evaluation criteria.
• Explain why the selected tool and model fit the brief.
• Identify at least one limit that requires human review.

2. Preparation and safety boundary

• Work on duplicate or versioned files; never overwrite the only source asset.
• Use only assets you own or are permitted to upload and transform.
• Record the app, model, account-plan assumptions, and whether a feature is Beta or Premium.
• Longer is not automatically better; every clause should guide generation or review.

3. Guided workflow

Step 1: Write a one-sentence output goal

Capture the input, the changed setting, the result, and one observation. Avoid changing several controls at once because the comparison will no longer explain what caused the difference.

Step 2: Describe subject and action

Capture the input, the changed setting, the result, and one observation. Avoid changing several controls at once because the comparison will no longer explain what caused the difference.

Step 3: Add setting and material cues

Capture the input, the changed setting, the result, and one observation. Avoid changing several controls at once because the comparison will no longer explain what caused the difference.

Step 4: Specify framing and lighting

Capture the input, the changed setting, the result, and one observation. Avoid changing several controls at once because the comparison will no longer explain what caused the difference.

Step 5: State invariants and exclusions

Capture the input, the changed setting, the result, and one observation. Avoid changing several controls at once because the comparison will no longer explain what caused the difference.

Step 6: Change one variable per iteration

Capture the input, the changed setting, the result, and one observation. Avoid changing several controls at once because the comparison will no longer explain what caused the difference.

4. Prompt and decision log

Goal: [one deliverable and channel]
Subject: [object and action]
Setting/material: [environment and texture]
Composition/light: [framing and light direction]
Constraints: [must keep / must avoid]
Review: [three observable acceptance criteria]

Version prompts as P01, P02, and P03. For each generation, record the model, reference assets, aspect ratio, selected variation, defects, and reason for acceptance or rejection.

5. Practice assignment

Write exploration, controlled, and final prompts for one product image.

1. Write the acceptance criteria before opening the app.
2. Create a baseline and no more than three controlled iterations.
3. Keep one rejected result and annotate why it failed.
4. Export the working file and a channel-ready preview.

6. Quality checklist

7. Common failure modes

• Judging only the most attractive result instead of the brief.
• Changing several variables and losing a meaningful comparison.
• Flattening or exporting before structural defects are resolved.
• Assuming an AI label guarantees accuracy, rights, or production readiness.

8. Official references

• Adobe Firefly overview

9. Course navigation

Previous lesson · Next lesson · Course appendix