Text-to-video prompting: the 8-part formula

HappyHorse-1.0's official guidance recommends a specific prompt shape, and in practice the model rewards it consistently:

Subject + action + scene + camera movement + lighting + visual style + mood + platform format

You won't use all eight parts in every prompt. But when a generation goes wrong, the fix is almost always "one of the eight parts is missing or vague." This lesson works through each part, then builds two complete prompts from scratch.

The eight parts

1. Subject — who or what the shot is about. Be concrete: not "a woman" but "a silver-haired woman in a yellow raincoat." The model allocates detail where you spend words.
2. Action — what the subject does. This is the single most neglected part. A prompt without a verb produces a beautiful clip where nothing happens. One clear action per clip; a 5-second shot cannot hold three.
3. Scene — where and when. Environment words do double duty in HappyHorse because they also drive the audio: "rainy night market" generates rain and crowd murmur, not just wet pavement.
4. Camera movement — how the (virtual) camera behaves: static, slow push-in, tracking shot, orbit, crane up, handheld. Lesson 6 is a full vocabulary; until then, one movement per prompt.
5. Lighting — golden hour, hard noon sun, neon, candlelit, overcast. Lighting words are the strongest lever on the feel of a shot.
6. Visual style — cinematic photorealism, 35mm film, anime, watercolor, claymation, documentary. HappyHorse advertises 50+ styles; name one explicitly or you'll get the default glossy look.
7. Mood — serene, tense, joyful, melancholy. Mood affects pacing, color grading, and the generated soundtrack.
8. Platform format — "vertical short-form video" or "widescreen cinematic" — reinforces the aspect ratio you set in the UI and shifts editing rhythm accordingly.

Building a prompt, live

Start with a flat idea: "a chef cooking."

Pass 1 — subject + action:

A young chef with rolled-up sleeves flips vegetables in a flaming wok

Pass 2 — + scene + camera:

A young chef with rolled-up sleeves flips vegetables in a flaming wok, in a cramped late-night Hong Kong dai pai dong, slow push-in from across the counter

Pass 3 — + lighting + style + mood:

A young chef with rolled-up sleeves flips vegetables in a flaming wok, in a cramped late-night Hong Kong dai pai dong, slow push-in from across the counter, lit by a single overhead tungsten bulb and wok flames, cinematic photorealism with shallow depth of field, energetic and warm

Run pass 3 and the audio comes with it: sizzle, clatter, street noise. That's the same prompt a director would give — subject, blocking, setting, camera, light, tone.

A second example, vertical

A ginger cat leaps between rooftop ledges at dawn, low-angle tracking shot following the jump, soft pink morning light, Studio Ghibli-style animation, playful and free, vertical short-form video

Note the format tag at the end matching a 9:16 ratio — the composition changes to center the action for a phone screen.

Failure modes and fixes

Negative prompting

Most HappyHorse hosts expose a negative prompt field. Use it for recurring artifacts, not wishes: "extra fingers, warped text, flickering" works; "bad quality" does nothing. If a host has no negative field, fold constraints into the prompt itself ("empty street" rather than negative "crowd").

Twenty-four ready-to-run prompts following this formula are in the prompt library. Next lesson: driving the model from a still image instead of a blank page.