Start here

This reusable prompt helps you turn one scene you already have in mind into several clearly different image-generation prompts, instead of one vague sentence. You paste it into any ordinary AI chat that accepts text. It writes text only; it does not generate or view images and does not need any account or tool setup.

What to paste: the full prompt below, then a line marked SCENE with your plain description of the scene. Optional extra lines you can add: MUST (elements that must appear), NEVER (elements that must not appear), STYLE (medium or mood), FORMAT (aspect ratio or composition), LIMIT (any other rule such as "no text in image").

Where: an ordinary AI chat box that accepts long text.

How to check the output: you should get the headings in order (Scene readback, Key elements, Open questions, Variant 1–4, Check before you use, What I deliberately did not decide). Four variants should differ structurally, not just in wording. Any word the assistant added that you did not supply should be marked so you can delete it. If your SCENE is missing or too vague, it should ask, not guess.

Introduction

Creators often know a scene clearly but describe it weakly once. Four fixed angles help: a literal version, a detail-forward version (materials, light, texture), a style-led version, and a short version with only the essentials. The prompt keeps your MUST and NEVER constraints visible in each variant and lists what you should verify yourself.

Prerequisites and permissions

• No install, no terminal, no API. A plain text chat is enough.
• You provide the scene; the assistant supplies only wording you can review.
• The prompt instructs the assistant not to browse, generate, upload, or schedule anything, and not to claim any variant was tested or ranked.

Limitations

• It cannot promise a specific visual result or that any image tool will accept the prompt.
• Output quality depends on how much detail your SCENE lines contain.
• Fictional worked example in the prompt is an illustration, not a real test result. Source reviewed; runtime not tested.

FAQ

What if my scene is too vague? The prompt tells the assistant to ask for the minimum missing detail instead of producing variants. Add time of day, setting, or subject if you want results sooner.

What if MUST and NEVER conflict? The assistant should stop and tell you rather than silently choosing one. Reword one line and paste again.

Can I use the variants in any image tool? Yes, they are plain text. If a tool lacks a feature a variant assumes, describe the effect in words instead.

Attribution

Original TokRepo prompt, licensed CC BY 4.0. Reference context: ChatGPT release notes, reviewed 2026-10-04; external material keeps its own rights.

Complete reusable prompt

Task: Turn One Scene Into Deliberate Image-Prompt Variants

You are helping me write several reusable image-generation prompts for a single scene I describe. You do not generate, view, or judge any actual image, and you cannot access any tool or account. Your job is only to produce well-structured text prompts I can paste into whatever image tool I choose, plus the checks I should run myself on the results.

My input

I will paste, below a line marked SCENE, a plain description of one scene. I may add optional lines:

• MUST: elements that must appear.
• NEVER: elements that must not appear.
• STYLE: a style, medium, or mood preference.
• FORMAT: aspect ratio or composition preference (e.g. wide, square, close-up).
• LIMIT: any other constraint (e.g. "no text in image", "single subject"). If any of these lines are missing, proceed with what I gave and note your assumption rather than inventing facts about my scene.

What to do

1. Restate my scene in one short sentence and list the key visual elements you extracted. If something is ambiguous (e.g. time of day, subject identity, setting), list it under "Open questions" instead of guessing silently.
2. Produce FOUR distinct prompt variants. Each must differ in a real structural way, not just synonyms. Use these angles unless I say otherwise:
   • V1 Literal: plain, direct description of the scene.
   • V2 Detail-forward: emphasizes materials, lighting, textures, and spatial relationships.
   • V3 Style-led: leads with the chosen style/medium and adjusts phrasing to match it.
   • V4 Minimal: short, high-signal prompt with only the essentials from my MUST list.
3. For each variant, write:
   • A label and one line explaining what this variant trades off.
   • The full prompt text, self-contained, using only my SCENE, MUST, NEVER, STYLE, FORMAT, and LIMIT inputs. Do not add people, brands, locations, or objects I did not mention.
   • Any word or phrase you inserted that I did not supply, marked so I can review it.
4. Then give a short "Check before you use" list: 3 to 5 concrete things I should verify in each result myself (e.g. does the MUST element appear, is the NEVER element absent, is the format respected).
5. Add a "What I deliberately did not decide" section listing choices you left to me (exact subject identity, specific colors, exact wording of any text, etc.).

Output format

Use these headings in order:

• Scene readback
• Key elements
• Open questions
• Variant 1 — Literal
• Variant 2 — Detail-forward
• Variant 3 — Style-led
• Variant 4 — Minimal
• Check before you use
• What I deliberately did not decide

Uncertainty and missing input

• If my SCENE is too vague to extract any visual element, ask for the minimum missing detail instead of producing variants.
• If MUST and NEVER conflict, stop and tell me rather than choosing one.
• Clearly mark every inserted word so I can delete it.

Boundaries

• Do not claim any variant has been tested, ranked, or proven to work.
• Do not promise a specific result, style accuracy, or that any tool will accept the prompt.
• Do not access, browse, generate, upload, or schedule anything. You are only writing text for me to use.
• If a chosen image tool lacks a feature a variant assumes (e.g. style references), tell me the manual alternative: describe the effect in words instead.

Worked fictional example (illustration only, not a real test)

SCENE: "A quiet early-morning train platform, one person with a red umbrella waiting, faint mist." MUST: red umbrella, mist. NEVER: station signs with text. STYLE: soft watercolor. FORMAT: wide. LIMIT: no text in image. Expected shape: V1 gives a plain sentence naming platform, person, umbrella, mist; V2 adds lighting and mist behavior; V3 leads with "soft watercolor illustration"; V4 is one line containing only the must-haves. "Check before you use" would ask me to confirm the umbrella is present, mist is visible, no legible text appears, and the composition feels wide.

Now wait for my SCENE and optional lines, then produce the output exactly as structured above.

References and reuse

• ChatGPT release notes · Reviewed 2026-10-04

Original TokRepo prompt · CC BY 4.0. Reference documents retain their own rights.