Creative Space

Contenido

Provide: Product information/images, audience, substantiated benefits, target platform/aspect ratio, and a presenter photo or appearance description.

Receive: A presenter-led vertical product video with music, script and draft post copy; multi-shot ads may include alternative hooks.

Formats

  • Multi-shot UGC, default: Presenter and demonstration alternate; brisk pacing and a strong first-three-second hook, usually 30–45 seconds, for discovery/testing.
  • Single-camera pitch: One digital presenter, setting and camera, holding/wearing the product throughout, usually 30–60 seconds; suitable for product pages and straightforward sales introductions.

For faceless scene-based product ads or unboxing, use Product Video Studio.

Tools and references

  • storyboard planning: marketing structure, script and storyboard; spoken-script structure for the single-camera format.

  • Flova, required: presenter identity, presenter/demonstration footage, voice and music.

  • FFmpeg, required: edit, mix and export variants.

  • Captions (optional): If requested, offer a suitable HyperFrames, Remotion or FFmpeg post-production workflow.

  • Host reading tools as needed: extract product-page facts/specifications and genuine buyer reviews.

For multi-shot work read and . For single-camera work read .

Boundaries

  • Product photos define the actual product. Preserve proportions, color/material, structure, logo/text and condition in every shot; do not generate an approximate substitute. Request missing views.
  • Copy functions, specifications, effects and numbers from supplied evidence. Show comparisons only when supported, especially for beauty, health and weight-related claims.
  • AI presenters introduce/demonstrate, not pose as real customers. Do not invent personal experience, reviews, sales or repeat purchases. Review/“I used it” language requires supplied real experience/test evidence, attributed buyer reviews, or an actual creator's footage; otherwise use demonstrative language.
  • Confirm rights/consent for real photos and voices. Do not create celebrity/influencer likenesses or imitate their voices; clone only the consenting person's own voice.
  • Remind the user to apply platform AI labels for synthetic presenters/voice and check advertising claims under local rules.
  • Discover current platform duration, ratio and safe-area requirements rather than relying on fixed remembered specifications.

Workflow

Product/audience/presenter/format (approval) → presenter (approval) → script (approval) → shot/sentence plan (approval) → generation (approval) → edit → optional multi-hook versions → delivery.

1. Confirm brief — approval

Request multi-angle product images, approved facts/benefits, audience and use situation, platform and ratio. Default to an AI presenter: request a consented photo or age impression, manner, clothing and setting. Multi-shot alternatives are hands/voiceover or supplied live creator footage. Choose a Flova voice, consented own-voice clone, or actual recording. Ask for genuine experience/test/review evidence before using testimonial language.

Multi-shot defaults to 30 seconds, 9:16, optionally up to 45; ask whether to test hooks. Single-camera defaults to 30–60 seconds, 9:16; choose one fitting setting and product placement (held, worn, at chest level). Record facts, permissible benefits and evidence in docs/brief.md; confirm before continuing.

2. Presenter — approval

Multi-shot: generate/approve character views as the sole identity reference throughout; skip for hands-only. Single-camera: use one clean ecommerce-ready reference image, not a three-view sheet.

3. Script — approval

Multi-shot: select problem/solution, tutorial, before/after, recommendation list or review from the audience/benefits/evidence. Reviews require genuine experience. Start with the hook; use short conversational sentences, only one or two main benefits, and a clear CTA. Annotate beats and approximate timings. Single-camera: use the marketing structure and divide into complete spoken sentences. Obtain script approval.

4. Shot or sentence plan — approval

Multi-shot entries record beat, duration, presenter/demonstration type, visual action, camera movement, references and speech. Alternate presenter and demonstrations; demonstrations must substantiate the benefit. For single-camera segments, record exact sentence, duration and small hand action. Confirm before generation.

5. Generate — approval

Choose mode/models through Flova Skill, usually Quick because FFmpeg handles the edit. Multi-shot presenter clips always reference the approved character sheet and reuse the description verbatim. Generate speech natively in a capable video model or combine voice with lip-sync; demonstrations reference the actual product image. Test one presenter and one demonstration clip before batching. Inspect identity, product, lips and unwanted subtitles; exclude generated music.

Single-camera: create the shared composite keyframe, whole-script voice and sentence-level lip-sync per its reference. Regenerate only failures; rewrite blocked descriptions, then switch models after repeated failure and tell the user.

6. Edit

Multi-shot uses mainly hard cuts, alternating presenter and demonstration. Speech is clearest, product/environment sounds secondary, music quietest. Single-camera follows the voice timeline as its sole timing basis. Export MP4 and watch the complete edit: first-three-second hook, supported benefits, clear CTA and synchronized lips and sound.

7. Alternative openings — optional multi-shot

Add two different hook types, changing only the first 3–5 seconds of footage/speech; reuse middle/ending exactly. Export separately and name by hook type.

8. Deliver

Provide MP4 and hook variants, script, storyboard/sentence plan, docs/brief.md and draft post copy; remind the user about AI labeling. Retain presenter references, footage, audio and assembly scripts so one sentence or hook can be replaced without rebuilding everything.

ugc-<product>/
├── source/      product images/facts, real experience, creator footage
├── docs/        brief.md, script, storyboard/sentence plan, post copy
├── presenter/   photo, character views/reference, fixed description
├── shots/       presenter, demonstration and pitch clips
├── audio/       voice, sentence slices, music
└── output/      final/variant MP4s and assembly scripts

From one photo to consistent character views

Multi-shot work needs front/side/rear appearance and full-body proportions, not merely a frontal photo; these reduce identity changes during turns, movement and framing changes.

Input and generation

Use a clear, evenly lit frontal photo, preferably waist-up/full-body, with consent for a real person. Without a photo, first generate/approve a frontal identity from description. Keep the photo's clothing or choose appropriate everyday clothing; specify cut/color and retain it throughout.

With a reference-capable image model and role reference, create the same person's full-body front, side and rear views, side by side on plain light gray, in natural stance and soft even light. An optional face close-up may supplement them. Use a 16:9 sheet, not an extra-wide strip; some video models limit reference ratios to 2.5. Split wider layouts into three images. Reuse the identity lock verbatim: “Keep this person's face, skin tone, hairstyle and build identical to the reference photo; do not change identity.”

Inspect — approval

Check face shape, hairline/style, skin/build across all views; plausible sides/back without invented patterns; resemblance to the original; and undistorted hands/feet. Obtain user approval. Regenerate a mismatch; after two unsuccessful corrections, request a clearer photo.

Reuse

Pass the character sheet to every presenter shot and state that it controls face, hair, build and clothing. Fix a written age impression/hairstyle/clothing/manner description and reuse it verbatim. For speech, either quote exact dialogue in a video model supporting speech, or pass a keyframe as first_frame and voice as driving_audio to a lip-sync model.

Single-camera product pitch

Use Flova Skill's general request structure, with these specialized instructions taking precedence; actual model parameters come from capability discovery. One digital presenter, one setting/camera, and one shared keyframe introduce the product throughout. Aim for realistic phone-front-camera footage, without cinematic lighting, grading or dramatic framing. Use Chinese prompts for a Chinese-speaking user; spoken copy follows the output language.

Establish the elements — approval

  • Presenter reference: Preserve identifiable features/manner from a supplied photo, making one clean ecommerce-ready reference. Otherwise describe age, build, distinctive features, hair and clothing. Use realistic photography, natural skin and real texture, not heavy beauty filters or a generic influencer face.
  • Product: Use the actual supplied image; request missing photos/views rather than generating a similar product.
  • Setting: One setting fitting presenter and product. Confirm all elements first.

Shared composite keyframe

Reference presenter, product and setting. Use a clear, large face in a medium close-up/bust shot, not full-body/wide framing that weakens lip-sync. The held/worn product shares the foreground with the face; shape/label must be readable. Tighten framing or bring hands/product closer if too small. Keep one composition, handling arrangement and lighting; explicitly locate person/product, without angle changes or insert close-ups. Default to 9:16, natural/interior light and slight handheld feel.

Script, segmentation and voice

Use a conversational marketing structure: opening pain/result, supported benefits, then purchase reason and CTA. Split only at complete sentence boundaries and the lip-sync model's actual audio limit; do not add shots for pacing or hard-cut by duration. Express benefit changes through speech and small gestures. Record segment ID, exact text, duration and hand action.

Use the user's recording if available. Otherwise generate a voice with explicit characteristics, speed and emotion. Default to generating the entire script once, then cut at inter-sentence pauses to preserve voice/tone. If generating sentence by sentence, verify voice and level consistency.

Video requests

Pass the same keyframe and each segment's voice to an audio-driven lip-sync model. The keyframe defines appearance; describe only changes/motion, using simple clauses and an explicit speaking state (natural pitch, enthusiastic explanation, smiling while talking).

Recommended structure: camera movement + emotion + speaking state + specific body action + optional background change. Example: “The camera stays nearly fixed with slight handheld movement. The smiling presenter explains the product in a friendly, clear rhythm. During the second sentence she raises the flask to chest height, turns its label toward the camera, then nods gently. The background remains quiet.”

Keep gestures small and tied to the benefit; the product stays visible. Include no music, no subtitles, and exact spoken text when the tool supports text alignment.

Assembly and inspection

Assemble in sentence order against the voice timeline, cutting at sentence boundaries. Remove generated clip audio and use the voice track. Do not retime video or fade speech out; put quiet music on a separate track.

Check:

  • Voice matches approved wording, complete/unclipped sentences, language, voice/tone and segment in/out points.
  • Keyframe has a clear large face, readable product, natural relationship and one setting.
  • Video lips match the slice; identity/product remain consistent without deformation/disappearance; one camera, continuously visible product, no unwanted music.
  • Whole video matches 9:16 or agreed format; color, lighting, face/product continuity and audiovisual synchronization hold.

Redo only affected segments. For model blocking, unsupported entitlement or service failure, choose another lip-sync model with the same keyframe/voice and tell the user which model changed.

Multi-shot UGC script and footage

Aim for creator-made recommendations: immediate hook, natural speech and visible product use rather than a polished brand commercial.

Choose a type

TypeStructurePresenter
Problem/solutionFamiliar frustration → use product → supported improvementRequired
TutorialStep-by-step demonstrationHands alone may suffice
Before/afterSupported before/after or comparison with ordinary methodHands may suffice
Recommendation listOne of three useful items, or three reasonsRequired
ReviewStrengths, honest limitations, conclusionRequired; genuine experience only

Without real experience material, choose a different type instead of fabricating a review.

Structure, 30–45 seconds

TimeBeatJob
0–3 sHookStriking visual plus a line that stops scrolling
3–10 sProblem/interestRecognize the viewer's situation or introduce the promise
10–25 sDemonstrationShow actual handling/use and what it looks like
25–38 sEvidenceSupported results, details or comparison
Last 3–5 sCTAClear next action: product cart, profile link or comments

Offer new information about every five seconds and do not end without a CTA.

Hooks and script

The first sentence is the hook, not self-introduction. Options: pain (“Still struggling with …? Watch this”); contrast (extreme but supported before/after in frame one); suspense (“I doubted it until …” / “Watch to the end”); numbers (“Three reasons …” / a documented price comparison); or direct demonstration of the strongest moment immediately. All claims require evidence: sold-out/repeat-purchase language needs actual data or experience.

Speak to one person as “you,” like a friend's recommendation rather than an announcer. Keep sentences roughly ten words and easy to say. One video has one main point and one or two benefits; copy facts and numbers from product materials. Deliver hook → beat-by-beat speech → CTA with approximate sentence timings.

Presenter routes and continuity

RouteMethodTypical uses
A. AI presenter, defaultPhoto/description → approved character views → all presenter shotsProblem/solution, lists, demos, comparisons
B. Hands/voiceoverHands/product footage with voice trackTutorials, comparisons
C. Real creatorUse supplied creator footage; write script, generate demo inserts, editReviews, genuine experience

Create/approve views per , then pass them to every AI presenter shot. Fix the written age/hair/clothing/expression/environment description and reuse it exactly. Generate speech natively for natural delivery, or create voice then lip-sync from keyframe/voice for tighter verbal synchronization. AI presenters demonstrate rather than claim personal experience.

Footage and realism

Use 9:16 with subject central or in the upper third, hook space above, and safe areas clear of bottom/right controls. Presenter footage should feel like phone-front-camera capture in a bedroom, kitchen, car or bathroom: natural light and slight handheld motion. Demonstrations show hands, product and action together; product stays clear and conforms to the facts locked in SKILL.md.

Avoid perfect studio light, seamless backgrounds and slider/gimbal smoothness. Favor hard cuts alternating presenter/demo, typically 1–3 seconds per shot; allow longer demonstrations to preserve the whole action.

Sound and testing

Speech is clearest, product/environment sound secondary, music lightest. Generate no music inside individual clips; add one consistent track in the edit.

To test hooks, typically make three versions with distinct types (pain, contrast, direct demo). Change only the first 3–5 seconds' visual and hook sentence; middle/end remain identical. Name files by hook, e.g. ad-hook-pain.mp4, ad-hook-contrast.mp4, so subsequent ad results can be compared.

Storyboard and asset handoff

When new shot planning is needed, the host uses this workflow directly; reuse a clear existing shot list or edit brief. Record each S01 segment’s script link, visual/information goal, duration/time, subject/action, camera or element motion, exact text/sound, transition and production method. Link E01 target elements to user material and actual resources; keep REF-E01 source identities separate and mark unavailable assets as pending. Check identity, space, action and sound continuity; retime to actual speech. For marketing, label applicable hook/turn/proof/cta roles and the viewer gain, using grounded demonstrations; teaching and narrative work retain their own structure. For an unclear expression, find a real reference and record what to borrow/replace. No separately named planning skill or file is needed.

Proceso creativo

  1. Flova

    Flova

    Generate creator footage

    Create presenter explanations, product demonstrations, and voiceovers.

  2. FFmpeg

    FFmpeg

    Assemble the video

    Alternate explanation and demonstration shots, add audio, and export.