Phantra AI: UGC Ad
Creates phone-shot product ads featuring a provided person and product image, using only dialogue and claims the person supplies.
UGC-style ad
A UGC (user-generated content) ad looks like one person filmed themselves on
a phone talking about a thing they use. In Phantra that person is
the one you are talking to, or someone whose face they have the right to
use, and the thing is their own product: both are elements, and the scene is
held to those photos. The words are theirs too. You write the prompt here
(free), save it with write_scene, then price it (price-and-render).
The two elements it needs
elements: find a character (the face) and a prop (the product). The ad renders against both; nothing is made up from a name.- Missing one? Ask the person for the picture.
add_elementtakes the https address they gave you of an image file or of the product's own page (kindcharacter or prop, aname, a few words ofnotes). A photo on their phone comes in throughupload_linkwithelement_kindandelement_name: give them theurl, they pick the file, then callelementsagain. Never find or guess a picture for them. - Roles on
write_scene: the face issubject, the product isproduct. The face ref goes FIRST; the render anchors on it. The pack shot goes second; a room or mood photo after, asplace,lightorpalette.
Whose words these are
Every line spoken, every claim ("it cleared my skin in a week"), every number and every review comes from the person, in this conversation. Ask what they want said and what they can stand behind, and use their words. Never write a testimonial, a star rating, a "customers say", a before-and-after or a statistic they did not give you, and never put a line in the mouth of a customer, a friend or an expert. A claim you cannot source to them is left out, and you say why.
The style block (phone feel, in craft terms)
Write it as cinematography, never as a brand or a creator's name: a phone's front camera at arm's length, face or chest height, wide with slight edge distortion; the room's own light (a window, a lamp, a mirror strip) with no fill; handheld with small natural sway, no stabiliser gloss; true-to-life colour, mild contrast, phone sharpness, a little noise in the shadows; a real room soft behind them. Realism comes from the light and the sway, not from roughness for its own sake.
The shots (timed, with to-camera moments)
Two or more windows, each ONE shot of 2-10 s, contiguous from 0 to the total (4-30 s; 15 is a good ad length). Mix to-camera and product moments:
- a to-camera open, mid-sentence, with the hook in their words;
- the product raised into frame, label to lens, then used (opened, applied, worn), hands and face in the same shot;
- a to-camera close with the one line they want remembered, cut abruptly.
Write what they say inside the shot ("Sam, to camera: '...'"). The voice is diegetic: "No background music. Only diegetic sound: their voice, the room, the product's own sounds."
A worked skeleton (replace everything in it; the quoted lines are the person's, taken down from this conversation):
Ultra-realistic UGC-style ad in 9:16. The attached photo of Sam is her exact
face and clothing; the attached photo of the tin is the exact product and
label. Style: front phone camera at arm's length, face height, wide with
slight edge distortion; bathroom light from a mirror strip, no fill;
handheld with small sway; true colour, mild contrast, phone sharpness, light
shadow noise; the bathroom soft behind her.
(0-4s) Sam, mid-sentence to camera: "okay so I've been using this for two
weeks" -- she lifts the tin into frame, label to the lens.
(4-10s) Same setup, closer: she opens it, scoops with two fingers and works
it into the back of one hand, glancing up at the lens.
(10-15s) To camera, tin at her shoulder: "that's it, that's the whole
routine." End abruptly.
No background music. Only diegetic sound: her voice, the tin lid, a tap
dripping.
Avoid: plastic AI sheen, CGI-looking skin, cartoon reactions, slow motion,
added captions or logos, named brands or people.
Save, price and render
Show the prompt, take their edits, then write_scene with prompt,
seconds, refs (face first), roles, title, and project_id when it
belongs to a project. Then quote the scene_id: it prices the keyframes
(stills x each) and the clip (one render per shot, each with credits
and a token; a window is fitted up to the model's shortest length) beside
the balance. Say the quote's numbers plainly -- the keyframes and their
credits, the shots and theirs, the balance -- and WAIT for a yes in this chat naming what to
run. Only then approve with what (keyframes or clip), the quote's
tokens and the same renderer choice; job (wait 15) until done;
scene shows the stills, the clips and the studio link. Tokens last an
hour and work once; an edit means a new quote and a new yes. Never approve
unasked or twice; refused for credits (insufficient_credits), stop and say
credits are added in the studio. Versions of this ad are ad-variations.
- Downloads
- Security audit
- Pass
- Last updated
- 2w ago
- Current version
- v1.0.0
- License
- MIT-0
