I Made an AI Influencer and Built Her a Full Commercial (Every Prompt)

High-level summary

• This video is a step-by-step tutorial showing how the creator built an AI “influencer” and a full cinematic commercial using one text-to-video model (Cling 3.0) inside the Higsfield platform.
• The host reads every prompt used, explains what each line does, and breaks down every shot of an ad made for his company (Growth School).
• Emphasis is on prompt design (how to describe characters, camera, lighting, delivery, and sound) and on practical Cling 3.0 + Higsfield features that make cinematic AI video possible.

Key product / technology features demonstrated

Cling 3.0 (AI video model)

• Generates up to ~15 seconds per generation.
• Native audio: dialogue, sound effects, and music are produced in the same generation (no separate dubbing).
• Character consistency across shots when using a reference / “elements” workflow.
• Multi-shot storyboarding: direct multiple camera cuts (up to six shots) in one prompt sequence with per-shot durations and audio.
• Multilanguage audio & lip sync (Hindi, Japanese, Spanish, Chinese, Korean, etc. — just write dialogue in the target language).
• Camera & cinematography-aware controls (camera moves, depth of field, speed ramps, slow motion, etc.).

Higsfield platform

• Elements library to upload/tag reference images for a consistent character (3–5 reference photos: front, side, smiling, neutral).
• Per-shot prompt boxes for multi-shot storyboarding with start/end times and individual soundscapes.
• Runs Cling 3.0 as the generator; annual plan offers a high generation allowance (host mentions “unlimited” on annual).

Concrete prompting rules, structures and vocabulary (actionable guide)

General prompt structure — include these components (in roughly this order):

1. Character description (wardrobe, age, prop, location).
2. Lighting line — e.g., “Warm afternoon light through window.” One sentence about lighting changes the mood.
3. Body action + quoted dialogue — words in quotes are spoken by the generated character (native audio & lip sync).
4. Delivery note — two-word pattern: mood + energy (e.g., “casual, curious”; “reflective, slightly tired”).
5. Camera direction — handheld, medium close-up, shallow depth of field, dolly tracking, crane down, low angle, Dutch angle, etc.
6. SFX: colon then list of sound elements (e.g., “SFX: traffic, distant chatter, footsteps on pavement”).
7. Optional: tag elements reference (e.g., “Arya reference image from elements”) at the start to force consistent face.

Important prompt rules and vocabulary

• Dialogue rule: put spoken lines inside quotes — the model will generate native voice and lip movement.
• SFX rule: use “SFX:” and list specific ambient sounds. Per-shot soundscapes allow different audio worlds across cuts.
• Character consistency: upload labeled reference images to the Elements library and tag the element in the prompt. If you skip tagging, the model will generate a random face.
• Camera vocabulary (compact list used directly in prompts): handheld, dolly, crane, arc, lateral, crash push. These words control camera behavior/motion.
• Multi-shot storyboarding:
  • Up to 6 shots per prompt sequence; maximum total ~15 seconds.
  • Minimum ~2 seconds per shot.
  • Each shot gets its own prompt, camera, and SFX; use logical shot progression (wide → medium → close).
• Action choreography: list physical moves in order and use speed modifiers (e.g., “slow motion as the kick connects; speed resumes on follow through”) for speed ramps.
• B-roll & macro: describe lens choice and framing (e.g., “extreme close-up, macro lens”) and add matching SFX for realism.
• Editing control: control where cuts happen by ending quoted dialogue at the desired cut point — the model generates the quoted audio exactly.

Blockquote note:

Dialogue inside quotes = generated native voice + lip sync. Use SFX: to define ambient audio for each shot.

Examples / tutorial items included in the video

• Full readout and line-by-line explanation of five interview-style prompts (examples include cafe conversation, street walking, office/Hindi prompt, tennis-court shoelace shot, classroom → stadium cut).
• Character-consistency walkthrough: how to create and tag an “element” (upload 3–5 images, label, tag in prompts).
• Camera vocabulary guide and examples (handheld, dolly tracking, crane down, low angle, Dutch angle, arc, lateral, crash push).
• Multi-shot storyboarding demo: up to six cuts with per-shot timings, per-shot audio worlds, and shot progression examples.
• Action/fight scene choreography prompt with speed ramp and per-movement SFX mapping.
• Multi-language prompting demonstration (Hindi example — write dialogue in the language and the model generates matching audio and lips).
• B-roll & macro technique (extreme close-up bee on flower with SFX).
• End-to-end ad build (Growth School commercial) showing eight different scenes and a four-shot cinematic car sequence told purely via SFX and shots (no dialogue).
• Practical tips: negative prompts for controlling unwanted features; iterative refinement (some shots took 2–3 tries); typical project time (the ad took roughly an hour from first prompt to finished ad).

Workflow & practical constraints

• One prompt generation can include video + synced audio + SFX + music — often replacing separate voice actors and post sound design.
• Character consistency requires tagging the saved element per generation. Without that tag, consistency is lost.
• Multi-shot limits: max 6 shots, min 2 seconds per shot, max total ~15 seconds.
• Some shots may require multiple generations (iterations of a few minutes each).
• Host provides copy-paste-ready prompts in a WhatsApp community and links them in the video description.

Analysis / takeaways

• Cling 3.0 + Higsfield translate higher-level cinematic direction (lighting, camera words, choreography, delivery shorthand, and SFX mapping) into usable, consistent short video outputs.
• The primary creative control now sits in precise prompt engineering: mood + delivery notes, body actions, camera vocabulary, and sound cues. Small specific details (lighting, props, micro-actions) drastically change perceived production value.
• The toolchain enables rapid prototyping across formats: talking-head interviews, multilingual spots, action choreography, B-roll macro shots, and multi-shot cinematic sequences — all from text and a few reference images.

Where to get the prompts / assets

• All prompts for the five interview prompts and the eight ad shots are available in the host’s WhatsApp community and linked in the video description.
• The entire ad was generated on Higsfield with Cling 3.0.

Speakers / sources

• Video host / creator: entrepreneur who has built startups and raised funding (subtitle reads “Seoia”; primary speaker reads and explains prompts). He is the source of the demonstrations and prompts.
• Technologies/platforms referenced: Cling 3.0 (AI video model) and Higsfield (platform and Elements library).
• The ad shown in the video is for the host’s company, Growth School.