Generate, refine, remix and transform prompts for ChatGPT, Gemini, Claude, Midjourney, Flux & more. The smarter way to work with AI.
Hand-picked prompts updated daily — copy, remix, or open in the builder.
A structured Image Generation prompt optimized for midjourney. Utilizes a 85mm optical perspective and Soft overcast diffused daylight through a large north-facing window to produce crisp, high-detail results without artificial smoothing.
A structured Image Generation prompt optimized for midjourney. Utilizes a 105mm optical perspective and Moody chiaroscuro single-source key light with deep velvety shadows to produce crisp, high-detail results without artificial smoothing.
A structured Image Generation prompt optimized for midjourney. Utilizes a 35mm optical perspective and Volumetric golden-hour rim light at 3800K with soft atmospheric haze to produce crisp, high-detail results without artificial smoothing.
Create a dark, emotional Instagram-style heartbreak memories photo edit while preserving the reference subject’s identity. This prompt combines a moody social-media profile interface, faded chats, cinematic lighting, film grain, and nostalgic breakup aesthetics for realistic Reel covers and social posts.
Surreal Fashion Editorial Collage 4 Worlds 1 Model: Instant copy-and-paste Image Generation prompt optimized for Midjourney. Easily adapt with dynamic parameters to match your exact project requirements. (Focus: When working with multi-frame grids, the AI's rendering engine has to manage a massive amount of vis...)
A structured Image Generation prompt optimized for midjourney. Utilizes a 50mm optical perspective and Warm candlelight ambient illumination at 2400K with gentle highlight rolloff to produce crisp, high-detail results without artificial smoothing.
A structured Image Generation prompt optimized for midjourney. Utilizes a 24mm optical perspective and Rembrandt 3-point studio lighting with 45° key light and soft fill bounce to produce crisp, high-detail results without artificial smoothing.
A structured Image Generation prompt optimized for midjourney. Utilizes a 24mm optical perspective and Soft overcast diffused daylight through a large north-facing window to produce crisp, high-detail results without artificial smoothing.
Technical system diagrams often turn into messy, unreadable clusters when generated by AI. Learn how to control layout grids, construct modular zones, and render crisp, production-grade architecture charts.
Vague quality buzzwords are ruining your AI portraits. Discover how to control depth of field, subsurface scattering, and specular reflections to render clean, high-contrast character art with zero synthetic gloss.
Create emotional Raksha Bandhan 2026 photos with your real face! Copy these premium, upgraded AI prompts with Strict Face Identity Lock for ChatGPT, Gemini, and Bing Image Creator to generate stunning traditional portraits, royal palace scenes, nostalgic childhood designs, and custom greeting cards in seconds.
A complete 2026 guide to AI-generated Independence Day portraits using the "Strict Face Identity Lock" technique. Covers step-by-step workflows, 20+ ready-to-use prompts across categories (studio portraits, boys, girls, couples, army tributes, name typography), troubleshooting common AI photo errors, and pro-tips for cinematic, face-accurate results on ChatGPT, Bing Image Creator, and Midjourney.
Browse our curated AI Image Prompt Studio — a dedicated gallery of 11,800+ curated prompts for Midjourney, Flux, DALL·E & Stable Diffusion. Filter by style, lighting and aspect ratio, then remix any prompt in one click.
From generation to refinement — build better prompts in seconds.
Interactive builder with dynamic variables. Text, dropdowns, sliders — create perfectly structured prompts every time.
Transform any prompt into Professional, Expert, Creative, Detailed, ChatGPT, Gemini, Claude versions instantly.
8 remix modes: Cinematic, Luxury, Viral, Dramatic, Minimal, Realistic and more. One click to transform.
Generate 5, 10 or 20 unique variations of any prompt. Never run out of creative ideas again.
Merge multiple prompts into a powerful unified prompt. Blend styles, themes and concepts seamlessly.
Generate surprising, creative prompts by category, tags and style. Perfect for creative exploration.
A production-ready methodology used by prompt engineers and AI professionals to consistently produce high-quality, reliable outputs from any large language model or image generator.
Every effective prompt begins by assigning an expert persona to the AI. Instead of asking
a generic question, frame the model as a domain authority: "Act as a senior UX researcher with 10 years
of B2B SaaS experience." This primes the model to draw on relevant knowledge patterns.
Context also includes the audience (who will read the output), the platform
(blog post vs. Slack message vs. code comment), and the domain (legal, medical, creative).
The richer the context, the narrower and more accurate the model's output space becomes.
The task definition must be atomic and unambiguous. Avoid stacking multiple objectives
into a single sentence. Use action verbs: Write, Summarise, Compare, Translate, Classify,
Rewrite, Extract.
Specify the exact deliverable format: a numbered list, a markdown table, a JSON object,
a paragraph under 150 words. Vague tasks produce vague outputs. The more precisely you describe
the endpoint, the closer the model will arrive to your expected result on the first attempt.
Constraints are the guardrails that prevent the model from drifting. They come in two forms:
positive constraints (use British English, use Oxford commas, maintain a warm but
authoritative tone) and negative constraints (do NOT use jargon, do NOT include
pricing information, avoid passive voice).
For image models, negative prompts (--no blur, watermark, text, low-quality) are
equally critical. Set explicit length limits, output formats, reading levels,
and any regulatory or brand restrictions.
Few-shot prompting is the single highest-leverage technique for production LLM workflows.
By providing 2–5 input/output pairs, you teach the model the exact pattern, tone,
and format you expect — without fine-tuning or any additional training cost.
Structure examples as Input → Output blocks, separated clearly. The model will extrapolate
the pattern to your actual query. Use this for classification tasks, entity extraction, structured
data generation, and style-matching scenarios where zero-shot prompts are inconsistent.
Each AI model has its own "prompt dialect." Use this reference table to write model-optimised prompts and avoid common cross-model mistakes.
| Technique / Parameter | ChatGPT / GPT-4o | Claude 3.5 / Sonnet | Google Gemini | Midjourney / Flux |
|---|---|---|---|---|
| Primary Prompt Structure | System + User messages. Use ### or --- delimiters to separate sections. Markdown renders natively in ChatGPT. |
XML structure tags: <context>, <instructions>, <examples>, <output>. Claude respects tag hierarchy precisely. |
Natural language with explicit grounding cues. Gemini responds well to structured bullet lists and numbered steps for multi-part tasks. | Comma-separated descriptors in order of priority: subject → style → lighting → camera → post-processing → parameters. |
| Reasoning / Chain-of-Thought | "Think step by step." "Reason through this before answering." Works with GPT-4o for complex logic and multi-step planning tasks. | Claude performs extended thinking natively. Add <thinking> tags or use the API's thinking parameter for deliberate reasoning traces. |
"Think through each step" or enable Gemini's "Deep Research" mode for complex analytical queries with real-time information grounding. | Not applicable to image models. Iterative refinement via --seed + /vary commands serves a similar debugging function. |
| Output Formatting | Explicit: "Format as a markdown table with columns X, Y, Z." GPT-4o renders Markdown, LaTeX, and code blocks in the web UI. | Wrap desired structure in <output_format> tags. Claude is highly faithful to formatting instructions and rarely deviates when explicitly structured. |
Specify format in the final line of your prompt. Gemini supports JSON output mode via API (response_mime_type: application/json). |
Aspect ratio: --ar 16:9. Quality: --q 2. Stylize: --s 750. Version: --v 6.1. Chaos: --chaos 20. |
| Few-Shot Examples | Provide 2–5 user/assistant pairs in the messages array. The pattern is extremely reliable for classification and extraction tasks. | Include examples inside <examples><example> tags. Multiple examples significantly improve output consistency on structured tasks. |
Inline examples work well. Use clear "Input:" / "Output:" labelling. Gemini 1.5 Pro handles up to 1M token context, supporting large example sets. | Use /blend to combine reference images. Use --sref (style reference) and --cref (character reference) for few-shot visual pattern matching. |
| Multimodal / Vision | Attach images directly to the user message. Reference image content explicitly: "In the attached image, identify and list all UI elements." | Upload images via API or Claude.ai. Claude Sonnet excels at document parsing, table extraction, and diagram interpretation from image inputs. | Native multimodal: text, images, video, audio, PDFs in a single prompt. Add grounding: "Using only information visible in this image, answer…" | Image prompts: paste a URL before text descriptors, or use /describe to reverse-engineer a prompt from any uploaded image. |
| Negative Constraints | "Do NOT include…", "Avoid…", "Do not use…" placed at the END of the prompt carries more weight than placement at the start. | Claude follows negative instructions reliably. Use <constraints> tags for grouped prohibitions in complex system prompts. |
"Do not mention…", "Exclude…". For grounding tasks: "Do not use information outside the provided document." Gemini respects boundary instructions. | Negative prompt: --no blur, watermark, text, cropped, low quality, jpeg artifacts, ugly, deformed. Essential for image quality control. |
| Persona / Role Setting | System message: "You are [persona]…" Most effective when placed first in the system prompt. Persist across conversation turns. | System prompt with role inside <role> tags, or Claude.ai's "Custom Instructions". Role adherence is very high in Claude Sonnet. |
Gemini supports system instructions via API. Web UI: start message with "Act as…". Gemini Gems allow persistent custom personas. | Style persona via artist references: "in the style of Moebius", "reminiscent of Syd Mead", "painted by James Gurney". Weight with ::2. |
Production-level techniques for the two most high-impact prompt engineering domains: text-to-image generation and advanced LLM system prompt design.
Effective image prompting is a layered specification process, not a single sentence. Professional image prompt engineers use a consistent ordering of descriptors that image models are trained to prioritise from left to right.
--style raw in Midjourney v6 to disable the default
stylisation engine and get outputs closer to your exact textual description.
Building reliable agentic LLM workflows requires understanding how models process long-context instructions, how to implement Chain-of-Thought reasoning, and when to choose zero-shot vs. few-shot vs. fine-tuned approaches.
--chaos.
Everything you need to know about prompt engineering, our platform tools, and how AI models interpret your instructions.
View All FAQsblurry, low quality, watermark, text, signature, ugly, deformed, extra limbs,
bad anatomy, NSFW.
--no parameter:
a woman in a garden --no fence, text, watermark.
--chaos (0–100), which controls
how diverse the 4-image grid results are from each other on a single generation.
--ar [W]:[H] — Aspect ratio. Common values: --ar 16:9 (landscape),
--ar 9:16 (portrait/mobile), --ar 1:1 (square), --ar 4:5
(Instagram portrait), --ar 3:2 (DSLR standard).
--v [version] — Model version. Use --v 6.1 for the latest.
--s [0–1000] — Stylize. Controls how strongly Midjourney's aesthetic training
is applied. --s 0 = raw prompt adherence; --s 1000 = maximum
Midjourney style. Default is 100.
--chaos [0–100] — Variation between the 4 generated images. Higher = more diverse.
--seed [number] — Locks the random seed for reproducibility. Use the same seed
to iterate on a specific composition.
--no [terms] — Negative prompt. E.g. --no text, watermark, blur.
A samurai in heavy rain, cinematic lighting, 35mm
film grain, muted earth tones, shallow depth of field --ar 2:3 --v 6.1 --s 250
--no text, watermark
Join thousands of creators who use HowToWritePrompt.in to craft more powerful, creative, and effective AI prompts.
Get curated prompt guides, advanced AI prompt engineering tricks, and resources sent directly to your inbox. No spam, unsubscribe anytime.
Subscribe to native browser web push notifications to get real-time alerts whenever we publish new prompts, AI tools, or tutorials.