I Wanna Love You AI Prompt Guide (Text to Video)

Photo mode does not need a prompt: the routine is built in. Text mode does. This guide is for the second mode, and for the optional extra direction you can add to a photo render.

The short version: describe the subject, the setting and the motion, in that order, and stop. A prompt is a direction, not a screenplay.

What the model actually reads

The model receives one combined prompt. In text mode it is your direction plus a fixed prefix that asks for a 12-second, vertical, one-shot clip with no text or logos. In photo mode it is the trend prompt plus your extra direction, if you add one.

That means a few things:

• You do not need to repeat “vertical” or “12 seconds”; the prefix already says it.
• You should not ask for captions, subtitles or a logo; the prefix forbids them.
• Your words are weighed against the reference image, if you attached one. A direction that contradicts the photo (different hair, different clothes) will produce a compromise or an artifact.

The three-part structure

A useful prompt has three parts:

1. Subject. Who or what is in the shot.
2. Setting. Where it happens, at what time of day, with what light.
3. Motion. What changes during the twelve seconds.

Example:

A young woman in a black jacket standing on a rooftop at dusk, city lights behind her, slow push-in as she turns her head toward the camera and smiles.

That is enough. It gives the model a subject, a place, a light and one action.

What to leave out

• Camera brands and lens names. “Shot on ARRI” does not change the output; it just consumes prompt space.
• Negative prompts as paragraphs. The system already excludes text, captions and logos. A long list of “no this, no that” tends to confuse more than help.
• Multiple cuts. The render is one continuous shot. “Then it cuts to…” will not work.
• Lyrics or copyrighted character names. They can trigger moderation, and they do not describe motion.
• Vague mood words with no visual anchor. “Cinematic” is fine; “make it iconic” is not a direction.

Extra direction in photo mode

In photo mode, your direction is appended to the trend prompt. Use it for small adjustments:

• “Warmer light, like golden hour.”
• “Keep the outfit from the photo, no suit.”
• “Slower lean-back, more of a smug pause before the wave.”

Do not try to replace the routine. If you want a completely different scene, use text mode.

Why shorter often wins

A twelve-second clip has limited room for events. A prompt with five actions produces a clip where none of them land. A prompt with one clear action produces a clip that reads.

The same is true of visual detail: one strong setting beats four competing ones. “A neon-lit arcade at night” is a better direction than “a neon arcade, a beach, a forest and a bedroom”.

Prompt examples to adapt

A pet version with a different setting

A small dog wearing a red bandana sitting on a skateboard in a sunlit parking lot, slow orbit around the dog as it turns its head to follow the camera.

A fashion-style clip

A person in a long coat walking through a rain-soaked street at night, neon signs reflecting on the pavement, camera tracking backward at walking speed.

A quiet portrait

A close portrait of a person by a window in the late afternoon, soft light, they look up from a book and hold the camera’s gaze.

These are not trend prompts; they are examples of the structure. Subject, setting, motion.

Testing cheaply

Text mode has more variance than photo mode because there is no reference image anchoring the look. Test at 480p first. If the composition works, rerender at 720p for the post. Paying for 1080p on an untested prompt is the most common way to waste credits.

A failed render returns its credits automatically, so a moderation refusal or a provider failure does not cost you the render. But a successful render that you simply do not like is a paid result; that is why the cheap test matters.

The honest limits of prompting

A prompt cannot guarantee a specific face without a reference photo. It cannot guarantee readable text, precise hand positions, or a exact dance move. It can set a scene, a mood and a broad action, and the model fills in the rest.

If you need a specific person in the clip, use photo mode. If you need a specific scene, use text mode and accept that the model will interpret it. If you need both, attach a photo as the first frame and describe the scene around it.

For the full workflow, read how to make the video. For photo selection, read the photo guide.