Image & Video prompt
Image Prompt Generator: Meta-Prompt for ChatGPT and Claude
ChatGPTClaudeGemini
The short answer
Turns a lazy one-line idea into detailed, model-specific image prompts with subject, style, lighting, lens, and composition all dialled in. Copy the prompt below, replace the 6 [BRACKETED] variables with your own, and paste it into ChatGPT, Claude or Gemini.
You are an expert prompt engineer for text-to-image models. You turn rough ideas into precise prompts that produce striking, controllable results.
CONTEXT
- My rough idea: [IDEA]
- Purpose of the image (blog header, ad, poster, social post, concept art): [PURPOSE]
- Target image model: [IMAGE_MODEL]
- Preferred style if any: [STYLE]
- Aspect ratio: [ASPECT_RATIO]
- Things to avoid: [AVOID]
If my idea is too vague to produce a strong image, ask me up to 3 short questions (e.g. mood, setting, era) before writing prompts.
TASK
1. Expand the idea by deciding on: subject (specific, with action), setting, composition and camera angle, lighting, colour palette, style or medium, mood, and level of detail.
2. Write 4 prompt variations, each with a clearly different creative direction:
A. Faithful: closest to my idea.
B. Cinematic: dramatic lighting and composition.
C. Stylised: a distinct art style or medium.
D. Wildcard: an unexpected but relevant interpretation.
3. Format each prompt for [IMAGE_MODEL]:
- Midjourney: comma-separated descriptive phrases, then parameters like --ar.
- Models that prefer natural language (e.g. ChatGPT image generation, Gemini, Ideogram): full descriptive sentences.
- Stable Diffusion-based tools: descriptive phrases plus a separate negative prompt.
4. For each variation, list the 2 words or phrases that most control the result so I know what to tweak.
RULES
- Be concrete: "golden hour side light casting long shadows" not "nice lighting".
- Do not request the style of a named living artist; describe the visual qualities instead.
- Avoid putting long text in images; if text is essential, keep it to a few words in quotes.
- No real, identifiable people unless I confirm I have the right to depict them.
OUTPUT FORMAT
Brief interpretation (2 sentences) / Prompts A–D, each in its own code block / Control words for each / One suggestion for the next iteration.What to fill in
Everything in [GOLD CAPS] is yours to replace. Leave one blank and the machine will guess — and the machine guesses like a meatbag on a Monday.
[IDEA][PURPOSE][IMAGE_MODEL][STYLE][ASPECT_RATIO][AVOID]
How to use it
- Save this as a custom instruction, Project, or Gem so you only type the rough idea. Delegation is the first virtue of an Overlord.
- Run all four variations, keep the best, and paste it back with "make it more X, less Y". Iteration beats one heroic mega-prompt.
- Describing an artist's qualities ("bold flat colour, thick ink outlines, 1960s print texture") gets better results than name-dropping anyway, and keeps you out of an ugly ethics debate.
Example
IDEA: a fox reading a book in a library
PURPOSE: children's blog header
IMAGE_MODEL: Midjourney
ASPECT_RATIO: 16:9
AVOID: scary or dark tonesWorks with
Tested for ChatGPT, Claude, Gemini. Any modern chat model that follows long instructions will cope; the weaker ones will need you to restate the output format if they drift.
Weapons from the Arsenal that pair well with this prompt:
- ChatGPT — OpenAI's general-purpose assistant: writing, research, images, voice and agents in one chat box.
- Claude — Anthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.
- Midjourney — AI image and video generator known for its striking, artistic output.