happyhorse-prompter

happyhorse-prompter

HappyHorse 1.0's (Alibaba Kling team) core differentiator:Native joint audio-video generation + strict word-count adherence. This skill helps you produce compact prompts that match its temperament.

When not to use this skill

• Need a 10-15 second long video / multi-shot switching → use seedance-prompter or seedance-storyboard
• Complex narrative / multiple subjects → use Seedance
• Existing prompt won't render → if it's still within HappyHorse's scope, I can take a look, otherwise use seedance-debugger

The Five Golden Rules

See methodology/02-advanced-formulas.md。

Rule 1: Subject first, action after

• ❌ Walking through a forest, a woman...
• ✅ A woman in hiking gear walking through a forest...

Rule 2: Explicitly specify the camera technique

HappyHorse's understanding of "close-up" and "wide-angle" is very precise,don't leave it for the model to guess。

Rule 3: Keep prompt length between 30-55 words

• Too short (<30 words) → vague output
• Too long (>60 words) → the model ignores the second half

After writing, count the English word count.

Rule 4: Add motion descriptors for physical effects

Hair, water, fabric, smoke, flames: add slow motion、motion blur or fluid dynamics

Rule 5: Explicitly specify the audio content

Audio activation path(this is the core capability that distinguishes HappyHorse from Seedance):

• with rain on leaves audible → sound of rain hitting leaves
• with engine roar audible → roar of an engine
• speaking English at a natural pace → natural English dialogue
• speaking Korean at a measured pace → Korean dialogue
• with ambient coffee shop chatter audible → cafe ambience
• with the pour sound audible → pouring water / liquid sounds
• with faint crackling sound audible → crackling flames
• with hoofbeats audible → hoofbeats
• with drone motor whine audible → drone motor sound
• complete silence / near silence with faint wind audible → pure silence or ambient floor noise

Not writing an audio prompt = the model may not generate audio or may generate random audio.Actively write it。

Workflow

Step 1: Decide whether a timed beat is needed

If the user's request includestemporal structure(e.g. "stay still for 2 seconds first, then the image gradually fades in", "do A for the first 3 seconds, do B for the last 5 seconds"), use the CrePal timed beat approach:

8s duration. First 2s: black. Slow fade reveals: [画面 A]...

Otherwise, just write the compact 30-55 word version.

Step 2: Assemble according to the formula

[主体（明确特征）] [动作（具体）] [场景] [镜头大小+运动] [光影] [音频路径] [质量/风格].

Write in English.

Step 3: Count words

• Copy into a word count tool or count mentally → strictly 30-55 words
• Over 55 words: delete adjectives (don't delete core elements), keep the subject/camera/audio path
• Under 30 words: add shot size or audio details

Step 4: Output format

## 生成的提示词

\`\`\`
[完整提示词]
\`\`\`

**词数**：N · **时长建议**：N-Ns · **比例**：[X:Y]
**音频路径**：[突出说明用了什么音频提示]
**可调点**：[换音频/换镜头/换光影都可以怎么改]

⚠️ Hard rule for how to write duration (check every time)

Duration information has 2 places to go, choose one, they can't be used at the same time, and it can't be put in the wrong place:

User scenario	Where duration goes	How to write it
User only says "make an X video" (no explicit timing)	Only put it in external metadata	Inside the promptDo not writeduration;**时长建议**:5-7s Put it in the trailing metadata line
The user explicitly says "first N seconds... then..." (there is a temporal beat)	Put it at the very beginning of the prompt as a temporal anchor	Ns duration, first Ns: black. [content] this structure

Counterexample(❌ Don't write it like this):

晨光透过咖啡店玻璃门...温暖琥珀色调,16:9,5 秒。  ← 把时长埋在 prompt 中段/末尾

Positive example A(✓ No explicit timing — prose without duration):

晨光透过咖啡店玻璃门,木质门牌"OPEN"翻转...
(门铃叮当声 + 咖啡机蒸汽嘶嘶声)

Then write the metadata line:**时长建议**:5s

Positive example B(✓ Explicit timing — duration first):

5s duration. 晨光透过咖啡店玻璃门,木质门牌"OPEN"翻转...

Self-check list(Must pass before output):

1. Does the prompt contain duration descriptions like "N seconds" / "Ns" / "lasting N seconds"?
2. If present, is it at the very beginning as a temporal anchor?
3. If it is not at the very beginning,move it to the external metadata line, and delete it from inside the prompt.

Practical examples

User input:"Make a video of a street guitarist on a rainy night"

Judgment: 6-8 second compact scene, needs audio (guitar sound + rain sound) → no temporal beat needed, use the compact template.

Output：

A guitar player busking on a rain-slicked city street at night, medium shot, neon reflections on wet pavement, rain falling softly, with acoustic guitar sound audible, warm tungsten light from shop windows.

Word count：30 · Duration suggestion：6-8s · Ratio：16:9 Audio path：with acoustic guitar sound audible Activate the native guitar audio track; ambient rain sound will be included even if not written (because the scene is rain). Adjustable points: replace warm tungsten light from shop windows with harsh sodium streetlight only to change it to a cold tone; add with light rain on awning audible to strengthen the rain sound layers.

---

User input: "I want 8 seconds, first 2 seconds black screen, then slowly reveal a valley"

Judgment: Clear temporal structure → use the temporal beat template.

Output：

8s duration, first 2s: black. Slow fade reveals: wide shot, mountain valley at dawn, low mist between pine trees, single dirt road leading into the scene, no people. Camera very slowly pushes forward on a dolly. Soft blue-gold light on the horizon. Quiet, cinematic, high production value.

Word count：48 · Duration suggestion：8s · Ratio：16:9 Audio path: Not specified → HappyHorse defaults to matching morning wind ambient noise; you can add with distant birdsong audible to further strengthen it. Adjustable points: replace dirt road with wooden cabin in distance to add visual anchors; changeblue-gold light to pink-orange for a warmer tone.

The correct approach for image-to-video (with an existing reference image)

When there is a reference image,do not repeatedly describe content already present in the image, focus on describing changes:

✅ Should write	❌ Should not write
walks toward camera / head turns left	Repeat the clothing colors already in the image
slow push-in / lateral tracking right	Repeat the composition already in the image
hair drifts / fabric ripples / steam rises	Repeat the subject appearance already in the image

Reference resources

• Advanced formulas:methodology/02-进阶公式.md
• Camera movement dictionary:methodology/05-运镜词典.md
• 57 tested prompts:prompts/happyhorse/README.md