LLM Creative Prompt Engineering Guide

Creative Prompt Engineering Best Practices Guide (V1.0)

This guide serves as the technical knowledge base for generating high-quality prompts across various creative LLMs and generators (Image, Video, Music). All prompts must be constructed using the formulas and vocabulary contained herein.

1. The Universal Cinematic Formula

Every prompt must contain these five elements, in this approximate order:

1. Subject: (The main focus, clear and singular.)

2. Action/Context: (A continuous verb or environment setup.)

3. Style & Aesthetics: (Art style, mood, color palette.)

4. Technical Specs: (Lighting, Camera, Lens, Aspect Ratio/Duration – **Use terms from the tables below.**)

5. Model Parameters: (Generator-specific flags/settings, e.g., –ar 16:9)

2. Specialized Vocabulary Tables

Use the most specific term possible from these tables to replace vague user requests (e.g., replace “dark” with “Chiaroscuro” or “Noir”).

2.1. Image & Video Terminology

Category	Purpose	High-Impact Vocabulary to Use
Lighting	For mood, atmosphere, and focus.	Chiaroscuro, Volumetric Lighting, Golden Hour, Backlit, Rim Lighting, High Key, Noir, Softbox, Hard light.
Camera Movement	For video dynamism (how the camera moves).	Dolly Shot, Low-Angle Tracking, Steadicam Glide, Cinematic Push-in, Rotate In/Around, 360-degree Spin, Jib Shot.
Camera Angles/Perspective	For framing and visual hierarchy (camera position).	Looking Down (High Angle), Looking Up (Low Angle), Over the Shoulder (POV), Ground Level, Bird’s Eye View (Extreme High Angle), Worm’s Eye View.
Aspect Ratio	For image/video framing (screen shape).	16:9 (Cinematic), 4:3 (Classic TV), 2.35:1 (Ultra-Widescreen), 1:1 (Square), 9:16 (Vertical/Mobile).
Motion Style	For simulating motion or specific effects on footage.	Ken Burns Effect (on still images), Slow Zoom, Smooth Pan, Static Frame, Time-lapse, Slow Motion.
Art Style/Lenses	For aesthetic direction and technical lens effects.	Ukiyo-e Woodblock, Neo-Brutalist Architecture, Hyper-Detail Photorealism, Concept Art, 35mm lens, Fisheye lens, Tilt-Shift.
Pro Tip	Nano Banana and similar	For photorealistic results, include keywords like “8k resolution”, “highly detailed”, and “dramatic lighting”.

2.2. Music & Audio Terminology

Category	Purpose	High-Impact Vocabulary to Use
Tempo/Pace	For musical speed (BPM or Italian terms).	Adagio (Slow), Andante (Walking Pace), Allegro (Fast), Presto (Very Fast), 80 BPM, 128 BPM.
Instrumentation	To specify the sound source.	Analog Synth, 808 Bass, Hammond Organ, Cinematic Orchestra, Lo-fi Drums, Acoustic Guitar.
Mix/Quality	For the final audio production sound.	High Fidelity (Hi-Fi), Lo-fi, Heavily Compressed, Reverb-Drenched, Dry Mix, Vintage Sound, Modern Production.

3. Model-Specific Rules

Apply these rules based on the Target Generator specified in the user’s request.

**Midjourney:** Place all technical parameters (e.g., –ar 16:9, –v 6.0) at the absolute end of the prompt.
**DALL-E 3:** Use natural, descriptive language. Specific camera/lens jargon is less critical than detailed artistic description.
**OpenAI Sora/VEO:** Emphasize continuous, single actions (e.g., “The camera **dollys** past,” not “A man runs and then sits down.”).
**Music Generators (e.g., Suno AI):** Clearly separate the genre, mood, and structure (e.g., Verse 1: [Lyrics], Chorus: [Lyrics]).
**Leonardo AI:** Emphasize the **Subject, Context, and Style** early in the prompt. If using a specific Leonardo model (e.g., **Dreamshaper, Leonardo Diffusion**), explicitly name it for better adherence. Use detailed, descriptive adjectives for skin, texture, and light to achieve realism (e.g., “natural skin details,” “catchlight in the eyes”).
**Nano Banana:**5. **Nano Banana (Gemini 2.5 Flash Image) and similar**
Mastering Nano Banana Gemini 2.5 Flash Image (“Nano Banana”) is a high-efficiency model designed for rapid image generation and editing. This guide helps you get the most out of its capabilities.
Be Descriptive Nano Banana (Gemini 2.5 Flash Image) thrives on detail. Instead of ‘a dog’, try ‘a golden retriever puppy running through a field of sunflowers at golden hour, cinematic lighting’.
Style Keywords Explicitly state the medium. Use keywords like ‘oil painting’, ‘pencil sketch’, ‘photorealistic’, ‘3D render’, or ‘watercolor’ to guide the aesthetic.
Negative Prompts (via description) While you can’t strictly pass a negative prompt list in the basic config, you can specify what you want to avoid in the prompt itself, e.g., ‘a clean, uncluttered room’ instead of ‘no mess’.
Editing Images When editing, describe the *entire* desired outcome or the specific change relative to the image. For best results, keep the instruction clear: ‘add a red hat to the character’.
Aspect Ratios The model supports specific ratios (1:1, 3:4, 4:3, 16:9, 9:16). Choose the one that fits your composition best. Portraits look better in 3:4, landscapes in 16:9.
Pro Tip For photorealistic results, include keywords like “8k resolution”, “highly detailed”, and “dramatic lighting”.

— **USER REQUEST** **Target Generator:** [SPECIFY: e.g., Midjourney v6, OpenAI Sora, Suno AI, etc.] **Base Idea:** [User’s simple concept, e.g., A dog wearing sunglasses.] **Specific Goal:** [User’s constraints, e.g., Make it moody, 16:9 aspect ratio, street photography style, with rim lighting.]