Prompt anatomy is the answer to a different question than prompt formula. Formula tells you the order to write. Anatomy tells you what each part controls, so you can repair the right part after the first image.
TL;DR: diagnose by prompt part, not by taste
- Subject controls what must exist in the image.
- Context controls why the image is being made and what counts as usable.
- Composition controls crop, distance, hierarchy, and negative space.
- Style controls material, lighting, palette, and mood after the subject is stable.
- Reference handoff controls what an uploaded image must preserve.
- Output rules control ratio, text policy, watermark policy, and production fit.
Prompt anatomy map
Use this map when a result fails. The failure usually points to one weak prompt part, not to the entire prompt.
| Anatomy part | Controls | Failure when missing |
|---|---|---|
| Subject | Object, person, scene, product, interface, or character. | The model changes the main thing or adds the wrong subject. |
| Context | Channel, audience, purpose, and level of polish. | The result looks pretty but unusable for the job. |
| Composition | Camera, crop, distance, layout, safe area, and hierarchy. | The frame feels cluttered, cropped wrong, or hard to reuse. |
| Style | Realism, material, lighting, palette, era, and mood. | The result is generic or inconsistent across a series. |
| Reference handoff | Which details a reference image protects. | Identity, product shape, UI layout, or brand color drifts. |
| Output rules | Aspect ratio, no text, no watermark, transparency, or safe area. | The image needs manual cleanup before use. |
| Review check | The first pass/fail criterion after generation. | Every revision becomes a full rewrite. |
How to read a prompt before you rewrite it
Underline each part. If a part is missing, do not add more adjectives. Add the missing control. If every part exists, change the one that explains the visible failure.
A labeled prompt anatomy example
The prompt below is intentionally split into parts. It is easier to debug than one long paragraph because every sentence has a job.
- Subject: Handmade ceramic coffee dripper with visible glazed texture.
- Context: Product-page hero image and launch social crop.
- Composition: Centered object, 4:5 frame, warm off-white background, soft shadow.
- Style: Premium ecommerce realism, softbox lighting from upper left, subtle ceramic highlights.
- Reference handoff: If a reference is uploaded, preserve silhouette, rim shape, and glaze color; background may change.
- Output rules: No text, no watermark, no extra props, clean negative space.
- Review check: Inspect silhouette first, then material texture, then crop.
Reference handoff is its own anatomy part
A reference image is not a magic instruction. It should say what stays fixed and what can change. For portraits, the reference may control face identity while wardrobe, background, and campaign styling are allowed to move.
Missing part symptoms
| Symptom | Likely missing part | Repair |
|---|---|---|
| Subject changes between generations | Subject or reference handoff. | Rewrite the subject sentence and clarify reference role. |
| Image looks good but not useful | Context. | Name channel, audience, and production use. |
| Crop is unusable | Composition. | Specify ratio, camera distance, and safe area. |
| Style feels generic | Style controls. | Add palette, material, lighting, and brand tone. |
| Logo or UI structure drifts | Reference handoff. | Say what the uploaded image controls. |
| Text artifacts appear | Output rules. | State no text, placeholder only, or add text later. |
First-result diagnosis by anatomy
- If the wrong object appears, fix subject first.
- If the object is correct but the frame fails, fix composition.
- If the frame works but mood is flat, fix style controls.
- If identity drifts, fix reference handoff before model choice.
- If the image is almost usable but needs cleanup, fix output rules.
Worked anatomy repair
Weak prompt
Make a premium skincare ad with a bottle, soft light, modern style, high quality.
Anatomy-based repair
Premium skincare product-page hero image of a translucent amber serum bottle, centered 4:5 crop, clean cream background, softbox lighting from upper left, visible glass refraction, preserve bottle silhouette and label position from reference image, leave negative space above for later headline, no generated text, no watermark. Review bottle silhouette first.
Use prompt anatomy in Vogue AI
When you adapt a Vogue AI prompt-library example, mark the anatomy parts that already work, then change only the missing part. This is faster than rewriting the prompt from scratch after every generation.
Prompt anatomy checklist
- Does the prompt name the exact subject?
- Does it explain where the image will be used?
- Does it specify crop, camera distance, and safe area?
- Does style language come after subject and composition?
- Does the reference image have a clear handoff rule?
- Does the prompt include one first-result review check?
FAQ
What is prompt anatomy for AI images?
It is the set of parts that make a prompt controllable: subject, context, composition, style, reference handoff, output rules, and review check.
How is prompt anatomy different from a formula?
A formula gives writing order. Anatomy explains what each part controls and how to diagnose failures.
Which prompt part should I fix first?
Fix the part that explains the visible failure: subject, composition, style, reference handoff, or output rules.
Do long prompts always have better anatomy?
No. A long prompt can still miss context or output rules. Good anatomy is complete, not just long.
Where should reference images be mentioned?
In a separate handoff sentence that says what the reference controls and what can change.
Can I use this anatomy for Midjourney, GPT Image 2, or Nano Banana?
Yes. The same anatomy helps you prepare and diagnose prompts, even when model behavior differs.