📄 prompt-guide.md

← 返回目录

Video Generation — Prompt Guide

Techniques for building effective video prompts per mode. If the user provides specific prompt text, use it as-is — suggest enhancements only.

Prompt Formula by Mode

| Mode | Formula | Focus | |------|---------|-------| | t2v | Entity + Scene + Motion + Aesthetic + Stylization | Full description | | i2v | Motion + Camera Movement (+ Sound) | Movement only — image defines entity/scene/style | | kf2v | Brief transition description (optional) | Images define start/end | | r2v | character1/character2 + Action + Lines + Scene | Character identifiers required | | vace | Edit instruction for the function | Function-dependent |

t2v / i2v Prompt Enhancement

Enrich short prompts by addressing each dimension:

| Dimension | Keywords / Patterns | |-----------|-------------------| | Lighting | daylight, firelight, overcast, backlight, rim light, neon | | Shot size | close-up, medium close-up, wide-angle, extreme full shot | | Camera angle | eye level, high-angle, low-angle, over-the-shoulder, aerial | | Composition | center, left-heavy, right-heavy, symmetrical, balanced | | Lens | long-focus, ultra-wide-angle fisheye, medium-focus | | Tone | warm tones, cool tones, low saturation, high contrast | | Time | daytime, night, dawn, dusk, sunset | | Camera move | camera pushes in, camera pulls out, camera moves left, fixed camera, orbiting |

i2v Prompt Specifics

Since the image already defines entity, scene, and style, focus on:

1. Motion: Specify what moves and how — slowly raises her right hand not just moves 2. Camera: Explicit direction and speed — camera slowly pushes in 3. Static hold: If camera should not move — fixed camera, no camera movement 4. Speed modifiers: slowly, quickly, gently, violently, gradually

Multi-shot Structure

Required: shot_type: "multi" AND prompt_extend: true.

[Overall description — theme, mood, narrative style]
Shot 1 [0–Xs] [Shot type, camera, lighting. Action. Dialogue. Sound.]
Shot 2 [X–Ys] [Transition. New shot type. Camera move. Action.]
Shot 3 [Y–Zs] [Scene change. Resolution. Sound.]

Sound Description (wan2.5/2.6 with audio)

Add sound to prompts for models that support audio generation:

| Category | Pattern | |----------|---------| | Voice | [Character] says "[line]" in a [emotion] tone, [speed], [timbre] | | SFX | [Source] [action], producing a [onomatopoeia] sound in [environment] | | BGM | [Music style] background music, [mood] |

r2v Prompt Pattern

Use character1, character2, etc. matching reference_urls order:

[Scene setting]. character1 [action with posture/expression].
character2 [action/interaction]. character1 says "[line]".
[Camera movement]. [Sound description].

Visual Style Keywords

| Style | Keywords | |-------|----------| | Felt | felt style, small felt gnomes | | 3D cartoon | 3D cartoon style | | Pixel | pixel style, pixel character | | Stop-motion | puppet animation, claymation style | | Special FX | tilt-shift photography, time-lapse photography |

Default negative_prompt

low quality, blurry, distorted, watermark, static, frozen, jittery,
inconsistent lighting, flickering

prompt_extend Decision

| Condition | Value | |-----------|:-----:| | Short prompt (< 30 words) | true | | Detailed prompt (advanced formula) | false | | Multi-shot (shot_type: "multi") | true (required) | | Precise timing/composition control | false | | Exploratory / creative | true |