prompt

Model-agnostic prompt engineering techniques covering role anchoring, structured output, explicit constraints, Few-shot, Chain of Thought, self-verification, and common pitfalls, not tied to any specific tool.

Retrieval-Augmented Generation (RAG) prompt patterns covering answer-only-from-context, cite sources, handle no-answer, fuse multi-passage evidence, and flag conflicts — emphasizing traceability and hallucination control.

Autonomous-agent system-prompt patterns covering goal/role definition, tool declarations, step-by-step decision loops, guardrails & permissions, error handling, and reflection — emphasizing control and safety.

Model-agnostic text-to-image (Midjourney/Stable Diffusion/SDXL) prompt structure covering subject, style & artists, composition & lens, light & material, negative prompts, and common params.

Code-generation prompt patterns covering requirement & context handoff, language/framework/version constraints, I/O contracts, edge cases & tests, and acceptance criteria — emphasizing compilable, runnable output.

Prompt-writing reference for DeepSeek that distinguishes the reasoning model R1 (deepseek-reasoner) from the chat model V3 (deepseek-chat), covering R1 hard rules (no system prompt, instructions in user, no few-shot, temperature 0.6) and V3 system-lock, few-shot, and JSON/function calling.

Gemini prompt engineering reference covering structured delimiters (XML/Markdown tags), long-context anchoring, internal-thinking and date/grounding clauses, native multimodal (text/image/audio/video) input, and model selection (3.5 Flash / 3.1 Pro / 2.5 Pro 1M / 2.5 Flash).

Stable Diffusion and SDXL prompt templates: six-segment structure, token weights, negative prompts, LoRA, ControlNet, CFG, steps, sampler, VRAM planning.

Kling AI video-generation prompt reference covering camera moves (push-in/pull-out/pan/tilt/dolly/orbit/crane/tracking), the one-move-per-shot rule, the six-segment prompt structure, and API params (camera_control/aspect_ratio/duration/model).

Perplexity prompt techniques covering "write questions not keywords", Focus modes (Web/Academic/Pro Search/Research/Reddit/YouTube/Writing/Math/Finance), date & source constraints, format constraints, chained follow-ups, and citations.

NotebookLM prompt techniques covering source management (PDF/URL/transcript/spreadsheet), answer-only-from-uploaded-sources with inline citations, verification prompts (find contradictions/gaps/counterpoints), and Learning Guide & Studio (study guide/infographic/audio) generation.

ChatGPT prompt templates by task: set a role, force JSON with a schema, review code, and use Few-shot examples or Chain of Thought.

Claude prompt templates for long docs (200K), code analysis, system prompts, structured output and multi-turn methods.

Midjourney prompt cheatsheet: parameters (--ar, --v, --q, --niji, --no), style keywords, camera and lighting modifiers, weight syntax and prompt skeletons.

Prompt-design patterns and templates for ChatGPT: role assignment, task framing and step breakdown, output-format constraints, few-shot examples and chain-of-thought steering — with common office, coding and writing scenarios. For turning vague asks into consistent, controllable model output.