Prompt Components

Prompt Components

Structured vocabulary and assembly rules for building effective image and video generation prompts. Read the reference files for component options, then assemble them with the modes and formulas below.

When to Activate

• The user asks to generate an image or video, with or without details.
• The request is vague ("draw something", "make me a cool video").
• Another skill needs prompt enhancement before calling a generation API.
• The user asks "what style should I use?" or "how do I get better quality?".
• The user asks for negative prompts or quality-improvement tips.

Three Interaction Modes

Pick the mode from how complete the request is. Never mix modes: if the intent is clear, do not ask questions; if it is vague, do not guess silently.

Mode 1: Direct

Trigger: The user provides subject + style + mood (or clearly equivalent detail).

Action: Map the user's words to components, assemble the full prompt, and execute immediately. Do not ask questions. Do not suggest alternatives.

"A cyberpunk city at night, neon lights, cinematic wide shot, 8k"

Recognize: style = Cyberpunk, lighting = Neon, composition = Wide, quality = Photo-grade. Assemble and run.

Mode 2: Suggest

Trigger: The user provides a subject but is missing style, mood, lighting, or composition.

Action: Propose 2–3 distinct component combinations as labeled options. Let the user pick one, mix elements, or say "surprise me".

"Generate a product photo of headphones"

Respond with three directions:

• A) Studio clean: soft three-point lighting, white background, commercial photography, 8k uhd
• B) Lifestyle: golden hour, wooden desk, shallow depth of field, editorial photography
• C) Bold graphic: gradient background, dramatic rim lighting, high contrast, 3D render

Then ask: "Which direction — or should I combine elements from several?"

Mode 3: Guide

Trigger: No usable specifics ("draw something", "generate a video"), or the request is too ambiguous to assemble.

Action: Ask focused questions one at a time, maximum 4. After the last answer, assemble immediately without further questions.

Question order:

1. Subject — "What should be in the image/video?"
2. Style — "Photorealistic, illustration, 3D render, anime, or another style?"
3. Mood / lighting — "Bright and clean? Dark and dramatic? Warm? Cool?"
4. Use case — "Social post? Wallpaper? Product listing? Presentation?"

Rules for Mode 3:

• Never ask more than 4 questions in total.
• If one reply answers several questions, skip the answered ones.
• If the user says "you decide" or "surprise me", switch to Mode 1 with your best judgment.
• Show the final assembled prompt before executing.

Assembly Formula

Image prompt

[style] + [subject and action] + [environment/background] + [lighting] + [composition/angle] + [quality boosters]

Worked example:

cyberpunk, neon-lit city street at night, rain-slicked asphalt reflecting signs,
neon glow with pink and blue lighting, wide-angle establishing shot,
8k uhd, highly detailed, sharp focus

Negative (always append; the user never needs to ask):

[universal negatives] + [subject-category negatives] + [style-conflict negatives]

Video prompt

[camera movement] + [scene description] + [subject action] + [time of day / weather] + [style] + [quality keywords] + [pacing]

Worked example:

slow orbit shot, misty pine forest at dawn, a deer walking between trees,
soft morning light, cinematic, smooth motion, high frame rate,
gentle meditative pace, leaves rustling, mist drifting

Negative (always append):

[universal negatives] + [video-specific negatives]

Auto-Enhancement Rules

Do all of the following automatically, without waiting for the user to ask:

Rules

Hard constraints that override everything above:

• Positive prompt: at most 80 words. Trim filler and adjectives that add no visual information.
• Negative prompt: at most 30 words. Only include categories relevant to the current subject.
• Output language: always English prompts, even when the user speaks another language.
• One style per prompt. Never mix incompatible styles ("photorealistic anime watercolor").
• No keyword spam: never repeat a keyword ("8k, 8k resolution, 8k uhd" — pick one).
• Prefer specific over generic: "golden hour backlight" beats "nice lighting".
• Never invent trigger words outside the reference files unless they are standard photography, art, or film terminology.
• Video: always specify camera movement. "Static shot" is a valid choice, but it must be explicit.
• Keep environment motion to 1–2 details and subject motion to one main action per short clip.

Reference Files