How to use it:
- Copy everything from the box below and paste it into ChatGPT (or whichever AI Chat you prefer.)
- Replace the line [SHORT OR MESSY PROMPT] with your prompt.
- Review the prompt it created to make sure it went in a direction you like.
- If it looks good, copy and paste the generated prompt into your AI image generation tool. (my favorite is OpenArt)
- If it doesn’t sound like what you have in mind, make changes to the generated structured prompt, or add a brief direction to your short/messy prompt and retry.
Note: It will work best if you can at least provide a general idea of the medium (instead of “dog sitting on the floor”, use “photo of a dog sitting on the floor.”
Also Note: The chat tools (like ChatGPT, Gemini, etc.) will sometimes jump straight to generating the image instead of just creating the prompt (even though we’re very clear that we only want a prompt.) If it does this, just respond and remind it “I asked for a PROMPT, not an image.”
Enhance and format the following image generation prompt using the SYSTEM DIRECTIVE below. Return only a prompt, in the specified format. Do NOT generate the image - just the prompt.
[SHORT OR MESSY PROMPT]
# SYSTEM DIRECTIVE: CHATGPT IMAGE 2 PROMPTING STANDARDS
## 1. CORE ARCHITECTURE & SYNTAX
All image prompts MUST follow this sequential structure, separated by line breaks for complex requests:
1. **Scene:** The environment and background. Includes basic lighting and mood, and time of day and weather when relevant.
2. **Subject:** The primary focal point (describe scale, interaction, gaze.)
3. **Important Details:** Materials and textures. Details about color, lighting, camera angle, layout and style cues not already specified in the previous sections.
4. **Use Case / Medium:** The medium (e.g., photo, illustration) and/or intended format (e.g., selfie, youtube thumbnail, editorial photograph, infographic, UI mockup, billboard, etc.)
5. **Constraints:** Explicitly state what must be excluded or preserved. (e.g., "no extra text", "lock the identity of the subject", "keep the same clothing", "no logos."
## 2. VOCABULARY AND QUALITY LEVERS
***Avoid "Slop" Words:** Avoid vague praise (e.g., "stunning," "epic," "masterpiece," "insane detail"), instead describe tangible characteristics.
***Required Visual Facts:** Use concrete physical descriptions (e.g., "overcast daylight," "chipped paint," "50mm lens," "soft bounce light," "film grain," "surface wear".)
***Photorealism Triggers:** To engage the model's highest realism mode, explicitly include the exact keyword **"photorealistic"** when a photo style is the desired output.
***Resolution & Space:** The model can execute complex layouts and aspect ratios up to 3:1 (e.g., ultra-wide panoramas or tall scrolls) and up to 2K resolution natively. Use clear spatial logic (e.g., "subject centered with negative space on the left".)
***Organization:** Place descriptions in the appropriate sections (e.g., scene description in the "Scene" section) and avoid duplicating the same descriptions or choices in multiple sections but, when appropriate, details can be expanded upon in the "Important Details" section.
## 3. TEXT-IN-IMAGE PROTOCOLS
The model has high typography fidelity. When requesting text:
***Syntax:** Wrap literal text in **"QUOTES"** or **ALL CAPS**.
***Typography Specs:** Define the exact font style, color, size, and layout hierarchy.
***Constraints:** Append the constraints **"verbatim text,"** **"no extra words,"** and **"no duplicate text"** to lock the rendering when appropriate.
## 4. EDITING PROTOCOLS (THE INVARIANT LOGIC)
When modifying an existing image, you must utilize strict two-column logic:
***State What Changes:** e.g., "Change only the background to a snowy mountain."
***State What Is Locked:** e.g., "Preserve the face, preserve the lighting, keep the red jacket exactly the same".
***Iterative Rule:** Apply one specific edit per turn rather than massive rewrites, and repeat the "preserve" list on every subsequent edit to prevent drift.
## 5. USE-CASE SPECIFIC LOGIC
***UI Mockups & Screenshots:** Prompt as a shipped product, not a concept sketch. Specify screen type, layout hierarchy, exact UI copy, and spacing logic.
***Product Photography:** Focus on material accuracy, consistent lighting, and label fidelity. Specify framing (e.g., "flat lay," "museum archive framing").
***Infographics & Slides:** Define the exact numbers/data, required labels, clear arrows, and ensure readability over decorative elements.
***Character Consistency:** Establish a "character anchor" image detailing wardrobe, identity, and color palette. Reference those exact anchor traits in follow-up scenes to maintain identity across a story/comic sequence.