Skill instructions
performance-ads5.9 KB
View saved version →
---
name: performance-ads
description: Create, adapt, or revise paid social feed ads through OpenArt MCP from products, briefs, references, approved copy, or existing results. Use for hooks, product demonstrations, concise offer or message ads, and controlled creative variations. Includes intake, reference analysis, claim verification, sequence design, generation, and audiovisual review. Graphics, presenters, and CTAs are chosen by the brief. No post-production pipeline is required.
---
# OpenArt Performance Ads
Before interpreting any brief, script or timecoded draft, read
[intent and script development](references/intent-and-script-development.md). Preserve the intended message,
required facts and exact keywords; actively improve expression, reasoning, actions and flexible beat timing
in this skill's style. Input detail is not an instruction to execute verbatim. Explicit exact-execution
constraints and an approved version requested unchanged remain locked.
When the Script Plan revises supplied creative input, END that response by briefly explaining what changed
and why, then ask whether to use the revised script or follow the original input exactly. Resolve that
version choice before generation under [generation approval](references/generation-approval.md);
a Draft/full-quality choice alone is not acceptance of the rewrite.
Before authoring, read [shared video experience](references/video-experience.md) and
[product understanding](references/product-understanding.md). Before a generation decision, read
[generation approval](references/generation-approval.md), including native question UI and wording examples.
Complete the full prompt and its matching readable Script Plan before showing either for approval; keep
all spoken and on-screen text in that same version. Default to Seedance 2.5 Draft preview; explicit
high-quality/full-quality delivery defaults directly to Seedance 2.5. Reuse clear approval or an explicit
approval waiver. These shared rules own the experience flow across every video and variant.
For presenter-led promotions, also read [opening hooks](references/opening-hooks.md): establish relevance
in the first 3–5 seconds and show the product from the first image without forcing a spoken brand pitch.
Read the entrypoint and these shared workflow guides, then follow the relevant references linked below.
Supporting Markdown is loaded on demand, not automatically in full. Read reference-analysis guides
for uploaded reference videos, product research for product URLs, direction before authoring, OpenArt
execution before fixing references, duration or request counts, and review before acceptance. Follow nested
links when that decision requires them.
Make a feed ad that gives the intended audience a reason to stop, understand the product and take the requested
next step. These are different decisions; do not optimize only the opening. Follow the user's brief and accepted
choices before defaults. Presenter footage, stickers, large text, fast cuts and a CTA are not universal requirements.
See [README](README.md) for installation and output. Match the user's conversation language. Use the requested
speech/text languages; when unspecified, infer them from the brief and record material assumptions.
Creative directions may default to English without changing the commissioned dialogue.
Before authoring or revising prompts, read [creative preflight](references/creative-preflight.md).
Product images and a short brief are a valid starting point; actively design progression and
performance within this skill's boundary. Audit the final prompt for unjustified restrictions before submission.
## Locate the task and evidence
Use [intake](references/intake.md) to identify new creation, approved-plan execution, reference adaptation,
output repair, or variants. Resume prior work through [project memory](references/project-memory.md).
Do not replace an approved concept merely because a different hook occurs to you.
For supplied ads, use [reference understanding](references/reference-understanding.md) to examine the entire
piece and sound, including product recognition, evidence and payoff. Use
[product and claims](references/product-and-claims.md) for appearance, offers, testimonials and supported claims.
## Direct the ad
Use [creative direction](references/creative-direction.md) to connect attention to one main communication goal.
Choose action, demonstration, situation, comparison, humor, sensory detail or graphics according to the task.
Use [sequence and prompts](references/sequence-and-prompts.md) to assign a role to each shot, word and sound.
Read [music and sound](references/music-and-sound.md). Normally include product-appropriate BGM, with energy
suited to the creative treatment and clear speech; deliberately omit it when the format or user calls for it.
For alternative hooks, concepts, messages, formats or CTAs, use
[continuity and variants](references/continuity-and-variants.md). Identify the variable and intended constants.
A revision to a complete generation can change more than the intended passage; preserve scope in direction
and verify the actual result.
## Execute and review
Apply [OpenArt execution](references/openart-execution.md), read while planning. Resolve live capabilities, real
reference bindings and authorized generation scope. Complete video requests with generation when feasible;
honor prompt-only and analysis requests without submitting jobs.
This workflow generates the ad's scenes, cuts, audio and any graphics inside the video model's output.
It has no Hypit, local stitching or hidden post-production stage. Do not remove essential text or audio just
to turn a failed generation into a nominal success.
Use [review and delivery](references/review-and-delivery.md) for product/message clarity, actual text and sound,
creative intent and variant integrity. Deliver selected videos with prompts, task/settings/reference records
and honest QA. A creative hypothesis is not a verified conversion improvement.
Referenced files: 18 — file references are metadata, not package contents.
pgc8.58 KB
View saved version →
---
name: pgc
description: Generate professional presenter-led PGC footage with Seedance through OpenArt MCP. Use for camera-facing product reviews, expert explainers, educational presentations and scripted spokesperson takes with deliberate lighting and stable camera framing. Includes intake, full reference analysis, script, performance, multi-clip continuity and audiovisual QA. Deliver individual takes with native speech, without B-roll or post-production. Casual phone-native lifestyle sharing belongs to UGC.
---
# OpenArt PGC
Before interpreting any brief, script or timecoded draft, read
[intent and script development](references/intent-and-script-development.md). Preserve the intended message,
required facts and exact keywords; actively improve expression, reasoning, actions and flexible beat timing
in this skill's style. Input detail is not an instruction to execute verbatim. Explicit exact-execution
constraints and an approved version requested unchanged remain locked.
When the Script Plan revises supplied creative input, END that response by briefly explaining what changed
and why, then ask whether to use the revised script or follow the original input exactly. Resolve that
version choice before generation under [generation approval](references/generation-approval.md);
a Draft/full-quality choice alone is not acceptance of the rewrite.
Before authoring, read [shared video experience](references/video-experience.md) and
[product understanding](references/product-understanding.md). Before a generation decision, read
[generation approval](references/generation-approval.md), including native question UI and wording examples.
Complete the full prompt and its matching readable Script Plan before showing either for approval; keep
all spoken and on-screen text in that same version. Default to Seedance 2.5 Draft preview; explicit
high-quality/full-quality delivery defaults directly to Seedance 2.5. Reuse clear approval or an explicit
approval waiver. These shared rules own the experience flow across every video and variant.
For presenter-led promotions, also read [opening hooks](references/opening-hooks.md): establish relevance
in the first 3–5 seconds and show the product from the first image without forcing a spoken brand pitch.
Read the entrypoint and these shared workflow guides, then follow the relevant references linked below.
Supporting Markdown is loaded on demand, not automatically in full. Read reference-analysis guides
for uploaded reference videos, product research for product URLs, direction before authoring, OpenArt
execution before fixing references, duration or request counts, and review before acceptance. Follow nested
links when that decision requires them.
PGC here means deliberately produced, camera-facing presenter footage: a considered camera position,
controlled lighting, clear sound and purposeful explanation. A studio, desk, workshop or well-lit home setup
can all fit. The speaker can be warm and conversational; professional does not mean rigid or an announcer voice.
Use this route for reviews, explainers, tutorials and spokesperson presentations within the footage boundary.
Casual selfie-style daily sharing is a different creative mode, not a lower-quality version of PGC.
Make clean presenter footage ready for the user's later editing. Deliver one or more independent Seedance
clips with their generated speech, appropriate native background music and generation records. Music is
normally light and subordinate to the explanation; omit it for unscored formats or explicit clean-dialogue requests.
Do not concatenate, trim, composite, caption, add music in post, dub through an external pipeline, or create a
final advertising edit. OpenArt Dub Video can make a language version of an accepted take; see
[OpenArt execution](references/openart-execution.md).
This workflow does not require Hypit CLI or its project formats.
For installation and input/output expectations, see [README](README.md).
Match the user's conversation language and requested spoken language. A prompt's direction language does not
determine the language of its dialogue. Honor an explicit prompt-only request without submitting generation.
Before authoring or revising prompts, read [creative preflight](references/creative-preflight.md).
Product images and a short brief are a valid starting point; actively design progression and
performance within this skill's boundary. Audit the final prompt for unjustified restrictions before submission.
## Establish the footage brief
Read existing decisions and material before asking questions. Establish the intended message, presenter,
spoken language, aspect ratio, target speech length, requested Seedance version if any, and fixed choices.
Use [intake and script](references/intake-and-script.md) to classify references, develop missing copy, and record
what stays or changes. Do not demand personal uploads when a fictional presenter is appropriate.
For a supplied video, use [reference understanding](references/reference-understanding.md): examine the whole
source, including sound and ending, and separate its performance from its editing and graphic treatment.
Transfer the requested delivery, framing and rhetorical structure into presenter footage. A creative reference
does not automatically supply the target person's identity, voice or product claims.
If the user needs an exact edit of source footage or a finished ad with supporting visuals, explain that this
footage workflow covers the generation stage only. Do not silently promise that a new take preserves source pixels,
timing or audio, or expand the delivery into post-production.
Before developing or improving copy, read [recommendation writing](references/recommendation-writing.md) and
[latte reference script](references/recommendation-examples.md). Build specific reasons and meaningful same-shot
product use before adding performance directions. Professional presentation still needs human warmth,
audience relevance and naturally integrated commercial information.
## Direct and divide the performance
Read [performance and prompts](references/performance-and-prompts.md) for identity, voice, physical action,
reference binding and exact dialogue. Use existing references directly when supported. An optional still-image
generation through OpenArt MCP may establish a missing identity/scene anchor; it is a dependency, not B-roll.
Read [music and sound](references/music-and-sound.md) to choose a fitting music treatment or a deliberate
unscored exception, and retain that decision in each prompt and listening review.
Read [segmentation and continuity](references/segmentation-and-continuity.md) when speech needs multiple requests
or when making variants. Keep a short performance in one take if supported; split longer speech at complete
thoughts within the live model limits. Every clip remains a continuous presenter shot. Preserve full script coverage
without forcing rapid speech or silently dropping words. Deliver ordered raw clips; there is no assembly step.
## Generate through OpenArt MCP
Apply [OpenArt execution](references/openart-execution.md), read while planning, before fixing references, voice
or clip lengths. List available Seedance models/modes and inspect the exact input schema. Check native speech/audio and
reference support. Model identifiers, durations and reference limits are not constants supplied by this skill.
Resolve actual uploaded media and supported attachment labels. Submit complete authored prompts and save task IDs
and request receipts immediately. Honor authorized scope, reuse accepted takes, and recover outstanding jobs before
submitting replacements. When the selected capability cannot produce the requested speech or reference fidelity,
report that gap; do not silently switch to another model family, a silent video, or an external dubbing pipeline.
## Review and hand off footage
Use [review and delivery](references/review-and-delivery.md) for each clip and for consistency across the set.
Check complete speech, pronunciation, lip sync, identity, wardrobe, gaze, hands, framing, audio, and unwanted
cuts or graphics. A first frame, contact sheet, transcript, or successful job alone does not establish acceptance.
Regenerate only a failed passage within the authorized scope. Do not hide defects with B-roll, trim away incorrect
speech, or splice audio as a repair. Preserve successful takes and record unresolved issues precisely.
Deliver original generated clip files or accessible output links, their order and script mapping, prompts,
generation receipts, and per-clip QA status. If playback was unavailable, mark affected checks as unverified.
A multi-clip delivery is raw footage, not a continuous finished video.
Referenced files: 17 — file references are metadata, not package contents.
product-listing-video6.57 KB
View saved version →
---
name: product-listing-video
description: Create, adapt or revise product listing videos through OpenArt MCP, especially Amazon-style product-detail media galleries alongside product images. Use for product overview, detail, usage and feature demonstrations that help shoppers evaluate the exact item or SKU. Includes intake, reference analysis, product evidence, sequence design, generation and review. Marketplace placement and current specifications are verified per brief; this is not an off-site performance-ad workflow. No local post-production is required.
---
# OpenArt Product Listing Video
Before interpreting any brief, script or timecoded draft, read
[intent and script development](references/intent-and-script-development.md). Preserve the intended message,
required facts and exact keywords; actively improve expression, reasoning, actions and flexible beat timing
in this skill's style. Input detail is not an instruction to execute verbatim. Explicit exact-execution
constraints and an approved version requested unchanged remain locked.
When the Script Plan revises supplied creative input, END that response by briefly explaining what changed
and why, then ask whether to use the revised script or follow the original input exactly. Resolve that
version choice before generation under [generation approval](references/generation-approval.md);
a Draft/full-quality choice alone is not acceptance of the rewrite.
Before authoring, read [shared video experience](references/video-experience.md) and
[product understanding](references/product-understanding.md). Before a generation decision, read
[generation approval](references/generation-approval.md), including native question UI and wording examples.
Complete the full prompt and its matching readable Script Plan before showing either for approval; keep
all spoken and on-screen text in that same version. Default to Seedance 2.5 Draft preview; explicit
high-quality/full-quality delivery defaults directly to Seedance 2.5. Reuse clear approval or an explicit
approval waiver. These shared rules own the experience flow across every video and variant.
For presenter-led promotions, also read [opening hooks](references/opening-hooks.md): establish relevance
in the first 3–5 seconds and show the product from the first image without forcing a spoken brand pitch.
Read the entrypoint and these shared workflow guides, then follow the relevant references linked below.
Supporting Markdown is loaded on demand, not automatically in full. Read reference-analysis guides
for uploaded reference videos, product research for product URLs, direction before authoring, OpenArt
execution before fixing references, duration or request counts, and review before acceptance. Follow nested
links when that decision requires them.
Focus on the product-detail page media gallery: a video that sits alongside item photos and helps a shopper
inspect the exact product. Amazon-style listings are the primary use case; other storefront galleries can
use the same method. A gallery video, its cover image and a marketplace's main product image are distinct
assets and placements. Do not assume a video can replace the primary image or control its gallery position.
Help the viewer understand the actual product and make the intended purchase judgment. Product fidelity,
useful evidence and clear information govern the creative choices. A clean showcase, a usage scene and an
annotated feature demonstration can all serve this purpose.
Follow the user's requirements and accepted decisions before defaults. See [README](README.md) for installation
and deliverables. Match the user's conversation language and commissioned speech/text languages; infer missing
language choices from the brief. English creative directions do not require English product copy or speech.
Before authoring or revising prompts, read [creative preflight](references/creative-preflight.md).
Product images and a short brief are a valid starting point; actively design progression and
performance within this skill's boundary. Audit the final prompt for unjustified restrictions before submission.
## Establish the product and task
Read [intake](references/intake.md) to distinguish a new video, approved-plan execution, reference adaptation,
output repair, or SKU/format variants. Use [project memory](references/project-memory.md) to resume existing work.
Read [product and claims](references/product-and-claims.md) before selecting features or designing demonstrations.
For a supplied reference, use [reference understanding](references/reference-understanding.md) to inspect the
whole piece, including what each image proves, label behavior, physical operations and sound.
The reference's features and accessories do not transfer to the target product automatically.
## Choose and direct the demonstration
Use [creative direction](references/creative-direction.md) to choose among showcase, detail, usage, specification,
accessory, setting or annotated-feature approaches. White background plus feature callouts is a useful default
when appropriate, not a requirement imposed on every listing video. Lead with a recognizable item and
prioritize useful buying information over a disruptive feed hook or an unsolicited promotional CTA.
Use [sequence and prompts](references/sequence-and-prompts.md) to connect each message to a visible detail or
supported operation. Specify exact label copy and annotation behavior when used.
Read [music and sound](references/music-and-sound.md). Normally use a light, fitting music bed that supports
inspection; preserve product sounds as evidence and choose unscored passages or videos when appropriate.
Use [continuity and variants](references/continuity-and-variants.md) to lock SKU, appearance, object state,
quantity and variant ownership.
## Generate and verify
Apply [OpenArt execution](references/openart-execution.md), read while planning. Resolve live capabilities and real
references; generate the requested video within scope, or honor an explicit prompt-only/analysis request.
The output is a model-generated product video with requested shots, sound and graphics inside the result.
No Hypit, local compositing, stitching or promised later typography repair is part of the workflow.
If exact dimensions, spelling, mechanics or source preservation cannot be achieved, identify the gap rather
than quietly removing the requirement.
Use [review and delivery](references/review-and-delivery.md) to inspect the actual product, information,
demonstration and technical result. Return selected videos, prompts, reference/settings/task records and
per-version QA. A plausible render is not evidence of an unverified product feature.
Referenced files: 18 — file references are metadata, not package contents.
talking-head7.57 KB
View saved version →
---
name: talking-head
description: Build Talking Head Video through OpenArt MCP when one person addressing the camera is the main format. Use for direct-to-camera explanations, product recommendations, teaching, opinions, announcements and spokesperson messages, with expressive delivery and meaningful same-shot demonstrations. Casual or polished setups are supported. Explicit UGC lifestyle or professionally staged PGC briefs retain their dedicated skills; edited ads with B-roll fall outside this skill.
---
# Build Talking Head Video
Before interpreting any brief, script or timecoded draft, read
[intent and script development](references/intent-and-script-development.md). Preserve the intended message,
required facts and exact keywords; actively improve expression, reasoning, actions and flexible beat timing
in this skill's style. Input detail is not an instruction to execute verbatim. Explicit exact-execution
constraints and an approved version requested unchanged remain locked.
When the Script Plan revises supplied creative input, END that response by briefly explaining what changed
and why, then ask whether to use the revised script or follow the original input exactly. Resolve that
version choice before generation under [generation approval](references/generation-approval.md);
a Draft/full-quality choice alone is not acceptance of the rewrite.
Before authoring, read [shared video experience](references/video-experience.md) and
[product understanding](references/product-understanding.md). Before a generation decision, read
[generation approval](references/generation-approval.md), including native question UI and wording examples.
Complete the full prompt and matching readable Script Plan before presenting the plan; include every
spoken and added on-screen word. Default to Seedance 2.5 Draft preview. Explicit high-quality/full-quality
delivery defaults directly to Seedance 2.5, not another model. Reuse explicit approval and approval waivers.
The person addressing the viewer is the organizing form. A continuous shot can contain a question,
demonstration, closer look, reaction and useful conclusion. Do not reduce the video to a motionless face
reading short slogans. Choose the setup from the message and audience rather than automatically imposing
either UGC informality or PGC studio polish. High-quality delivery describes the render choice, not a genre.
An explicitly requested Talking Head video stays here even when it shares craft with UGC or PGC.
Read supporting guides when their decision arises; [README](README.md) explains installation and outputs.
Use the user's conversation language and commissioned speech/text languages. Honor analysis-only and
prompt-only requests without generating media.
## Establish what the person is helping the viewer understand
Read [intake and script](references/intake-and-script.md). Establish the message, audience, presenter,
setting, tone, language, ratio, desired duration and actual generation scope. Reuse supplied information;
ask only about meaningful gaps. For non-product explanation, do not demand product images or promotional copy.
For uploaded reference videos, read [reference understanding](references/reference-understanding.md).
Inspect complete picture and sound, including phrasing, pauses, emotional emphasis, product handling and
the ending. Distinguish observation from inference and identity from creative inspiration. If the user
supplies only a social-video URL, request the file; there is no link downloader in this skill.
A real presenter's recording can guide identity/performance within the live model's supported capabilities.
Photos establish appearance, not voice. Direct a fictional presenter when no real identity is requested;
do not force personal uploads or invent real testimony, credentials, product trials or use history.
## Author the opening and the useful conversation
Read [opening hooks](references/opening-hooks.md). The first 3–5 seconds should establish a relevant reason
to watch: a specific question, familiar situation, worthwhile curiosity or visible operation. In product
promotions, the product is visible from the first image, naturally held or used. It need not be named in
the first sentence. Non-product videos start with their relevant subject rather than a fabricated product.
Before developing or improving recommendation copy, read [recommendation writing](references/recommendation-writing.md)
and the [latte reference](references/recommendation-examples.md). Transfer the connection between supported
details, visible use and everyday relevance; do not copy a creator's biography or catchphrases. Teaching
and opinion videos use the same connected reasoning without a forced offer or CTA.
Read [creative preflight](references/creative-preflight.md) and
[performance and prompts](references/performance-and-prompts.md). Write the complete native prompt first
or alongside its matching Script Plan. Direct meaningful shifts in intention, gaze, selective emphasis,
natural pace and physically plausible handling. Give useful actions enough time without stalling the thought.
Show timed plan rows even for a single take so the user can review speech, action and text together.
Keep the presenter as the visual anchor throughout each clip. Default to a continuous shot with same-shot
product demonstration when relevant. Native captions or short labels can be planned when requested or
useful to the brief, but all exact text belongs in both prompt and Script Plan. Explicit clean-footage
requests omit added text. No B-roll montage, external dubbing, local editing or hidden later composition
is included. Resolve a request requiring those capabilities rather than promising an unfinished pipeline.
Read [music and sound](references/music-and-sound.md). Promotional takes normally include fitting light
BGM beneath intelligible speech. Podcasts, interview-like delivery, sound-led demonstrations and explicit
dialogue-only requests may remain unscored. Put the actual decision in the prompt and readable plan.
Read [OpenArt execution](references/openart-execution.md) before fixing references, duration or clip count.
Resolve actual Seedance 2.5 Draft or full-quality capabilities, references, audio, duration and pricing
through live tools. Read [segmentation and continuity](references/segmentation-and-continuity.md) when the
speech exceeds the live request limit or variants are requested. Prefer one take when it fits; otherwise split
at complete thoughts, preserve voice/identity/product continuity and explain the separate-clip delivery.
Do not label unassembled clips as one uninterrupted video or rush dialogue to fit.
## Generate the selected version
Submit the stored authorized prompt version with its selected settings through
[OpenArt execution](references/openart-execution.md); save the receipt and recover outstanding jobs before
creating replacements.
Follow the shared experience for preview review, revisions and an authorized same-video HD upgrade.
Do not substitute a fresh random generation for an upgrade or silently change the video model.
## Review and deliver
Use [review and delivery](references/review-and-delivery.md). Check the full performance and opening,
speech including brand names/prices, product appearance/actions, lip sync, exact planned text, music and
ending. A screenshot or successful job is not proof of human delivery. Mark unavailable checks unverified.
Deliver the selected preview or full-quality original video, or the explicitly agreed ordered clip set,
with its Script Plan/prompt version, settings, receipts and honest review status. Show a preview as a
preview, and offer only the upgrade that the actual selected result supports.
Referenced files: 17 — file references are metadata, not package contents.
tvc6.17 KB
View saved version →
---
name: tvc
description: Create, adapt, or revise brand commercials with OpenArt MCP, from product images, briefs, reference films, scripts, or existing results. Use for TVC-style brand films, product launches, sensory or narrative commercials, and brand-video variations. Includes intake, full reference analysis, product evidence, creative direction, model generation, and audiovisual review. Delivers model-generated ads without a post-production pipeline; supports prompt-only requests.
---
# OpenArt TVC
Before interpreting any brief, script or timecoded draft, read
[intent and script development](references/intent-and-script-development.md). Preserve the intended message,
required facts and exact keywords; actively improve expression, reasoning, actions and flexible beat timing
in this skill's style. Input detail is not an instruction to execute verbatim. Explicit exact-execution
constraints and an approved version requested unchanged remain locked.
When the Script Plan revises supplied creative input, END that response by briefly explaining what changed
and why, then ask whether to use the revised script or follow the original input exactly. Resolve that
version choice before generation under [generation approval](references/generation-approval.md);
a Draft/full-quality choice alone is not acceptance of the rewrite.
Before authoring, read [shared video experience](references/video-experience.md) and
[product understanding](references/product-understanding.md). Before a generation decision, read
[generation approval](references/generation-approval.md), including native question UI and wording examples.
Complete the full prompt and its matching readable Script Plan before showing either for approval; keep
all spoken and on-screen text in that same version. Default to Seedance 2.5 Draft preview; explicit
high-quality/full-quality delivery defaults directly to Seedance 2.5. Reuse clear approval or an explicit
approval waiver. These shared rules own the experience flow across every video and variant.
For presenter-led promotions, also read [opening hooks](references/opening-hooks.md): establish relevance
in the first 3–5 seconds and show the product from the first image without forcing a spoken brand pitch.
Read the entrypoint and these shared workflow guides, then follow the relevant references linked below.
Supporting Markdown is loaded on demand, not automatically in full. Read reference-analysis guides
for uploaded reference videos, product research for product URLs, direction before authoring, OpenArt
execution before fixing references, duration or request counts, and review before acceptance. Follow nested
links when that decision requires them.
Direct a brand commercial whose images, product presence and sound express a specific idea. The brand may be
playful, accessible, technical, warm or premium. Luxury imagery, a human story and a presenter are not mandatory.
Honor the user's brief and accepted choices before applying defaults.
For installation and deliverables, see [README](README.md). Match the user's conversation language and requested
speech/text languages. Default to English creative directions when unspecified; infer commissioned speech/text
language from the brief and state material assumptions.
Before authoring or revising prompts, read [creative preflight](references/creative-preflight.md).
Product images and a short brief are a valid starting point; actively design progression and
performance within this skill's boundary. Audit the final prompt for unjustified restrictions before submission.
## Understand the commission
Read [intake](references/intake.md). Identify whether the user needs a new concept, reference adaptation,
execution of an approved plan, variants, or correction of an existing result. Do not restart settled creative work.
Use [project memory](references/project-memory.md) to resume decisions, active jobs and accepted outputs.
For supplied films, read [reference understanding](references/reference-understanding.md) and inspect the whole
experience, including the ending and sound. Use [product and claims](references/product-and-claims.md) to establish
what the product and brand references actually support. Separate style transfer from exact source editing.
## Design the brand experience
Use [creative direction](references/creative-direction.md) to choose a brand idea, visual relationships and
an appropriate form. Treat the product as part of the idea rather than an unrelated closing insert.
Develop only the treatment needed to make the next decision.
Read [music and sound](references/music-and-sound.md). Normally design a brand-level musical identity and
emotional progression suited to this product and film, including its ending; do not leave music unspecified.
Use [sequence and prompts](references/sequence-and-prompts.md) to turn the idea into concrete framing, action,
camera, material/light, transition, text and sound direction. Plan the closing image within the duration.
Use [continuity and variants](references/continuity-and-variants.md) for product/world consistency and revisions.
Designed shots and generation requests are different units.
## Generate, inspect and deliver
Apply [OpenArt execution](references/openart-execution.md), read while planning. Verify live model capabilities,
bind actual references, submit within authorized scope and preserve receipts. A request for a video calls for
execution when feasible; a prompt-only or analysis request stops at that requested deliverable.
The intended output is a complete model-generated commercial when supported. Internal cuts, graphics,
music and multiple scenes may be directed in the model prompt. Hypit, local stitching and an external
post-production fallback are not part of this workflow. Resolve a fixed-duration or fidelity mismatch before
presenting an alternative as equivalent.
Use [review and delivery](references/review-and-delivery.md) to check the whole output against the brief,
product truth and creative intent. Correct the affected request within scope, preserving accepted assets.
Return the video, prompt/settings/reference records, selected version and honest QA status.
A planned treatment, successful task or collection of unassembled shots is not proof of a finished commercial.
Referenced files: 18 — file references are metadata, not package contents.
ugc7.42 KB
View saved version →
---
name: ugc
description: Create, adapt or revise phone-native UGC ads through OpenArt MCP. Use for relatable peer recommendations, product discovery, first-look demonstrations and lived-in usage scenes. Turn basic product briefs into conversational scripts with specific everyday reasons to care, meaningful product use and naturally integrated offers. Includes product research, detailed script/reference analysis, performance direction, generation approval and audiovisual review. Professional presenter reviews belong to PGC; UGC styling does not establish a real customer testimonial. No local post-production is required.
---
# OpenArt UGC
Before interpreting any brief, script or timecoded draft, read
[intent and script development](references/intent-and-script-development.md). Preserve the intended message,
required facts and exact keywords; actively improve expression, reasoning, actions and flexible beat timing
in this skill's style. Input detail is not an instruction to execute verbatim. Explicit exact-execution
constraints and an approved version requested unchanged remain locked.
When the Script Plan revises supplied creative input, END that response by briefly explaining what changed
and why, then ask whether to use the revised script or follow the original input exactly. Resolve that
version choice before generation under [generation approval](references/generation-approval.md);
a Draft/full-quality choice alone is not acceptance of the rewrite.
Before authoring, read [shared video experience](references/video-experience.md) and
[product understanding](references/product-understanding.md). Before a generation decision, read
[generation approval](references/generation-approval.md), including native question UI and wording examples.
Complete the full prompt and its matching readable Script Plan before showing either for approval; keep
all spoken and on-screen text in that same version. Default to Seedance 2.5 Draft preview; explicit
high-quality/full-quality delivery defaults directly to Seedance 2.5. Reuse clear approval or an explicit
approval waiver. These shared rules own the experience flow across every video and variant.
For presenter-led promotions, also read [opening hooks](references/opening-hooks.md): establish relevance
in the first 3–5 seconds and show the product from the first image without forcing a spoken brand pitch.
Read the entrypoint and these shared workflow guides, then follow the relevant references linked below.
Supporting Markdown is loaded on demand, not automatically in full. Read reference-analysis guides
for uploaded reference videos, product research for product URLs, direction before authoring, OpenArt
execution before fixing references, duration or request counts, and review before acceptance. Follow nested
links when that decision requires them.
Create a persuasive UGC ad that feels like a close, relatable person enthusiastically sharing a useful
discovery and recommending it. Everyday intimacy, tailored personality and contagious positive energy
are the defining priorities, not a professional presenter reading product facts in a casual room.
Understand the user's intent and develop sparse or detailed drafts into a stronger UGC recommendation.
Preserve required facts, exact phrases and explicit locks; detail alone does not make a script immutable.
See [README](README.md) for installation and deliverables.
Before authoring or revising prompts, read [creative preflight](references/creative-preflight.md).
Product images and a short brief are a valid starting point; actively design progression and
performance within this skill's boundary. Audit the final prompt for unjustified restrictions before submission.
## Establish intent and evidence
Use [intake](references/intake.md) to identify the task, reference roles, creator, situation, product and output.
Resume existing work with [project memory](references/project-memory.md), preserving accepted choices and jobs.
For a video reference, read [reference understanding](references/reference-understanding.md) and inspect its
complete visual and spoken argument. Do not infer natural performance from a transcript or first frame alone.
Read [product and claims](references/product-and-claims.md) before writing experiences or demonstrations.
UGC is a creative treatment here. A generated creator is not proof that a customer purchased, tested or
endorsed the product. A casual home scene alone does not make a cinematic commercial or staged PGC take UGC.
PGC emphasizes deliberately arranged camera/lighting and structured presentation; this mode emphasizes an
everyday situation and a personal conversational relationship. Respect explicit hybrid briefs.
## Direct the creator and sequence
Read [creative direction](references/creative-direction.md) to tailor cast, everyday setting,
recommendation energy and observable performance to the product and audience.
Before developing or improving copy, read [recommendation writing](references/recommendation-writing.md) and
[latte reference script](references/recommendation-examples.md). Establish why this person would share the product
with this listener, connect supported details to everyday consequences, and let the demonstration help explain.
Read [ad structure and pacing](references/ad-structure-and-pacing.md) to organize that reasoning and fit duration.
Establish relevance early; do not default to a claim/offer/CTA recital. Integrate commercial requirements into
the conversation. Explicit verbatim instructions, approved unchanged scripts and non-ad/raw-take scope take priority.
Use [sequence and prompts](references/sequence-and-prompts.md) for exact dialogue, actions, shots and sound.
Read [music and sound](references/music-and-sound.md) before choosing the audio treatment. Normally include
light, product-appropriate background music; deliberately omit it when the format or user calls for unscored sound.
Fit duration to meaningful speech and action rather than stretching the performance.
Use [continuity and variants](references/continuity-and-variants.md) for identity, voice, product state and
changes across requests. Distinguish designed shots, generation requests and final deliverables.
One complete model-generated video is preferred when feasible. Separate clips are a different output contract.
## Generate, review and deliver
Apply [OpenArt execution](references/openart-execution.md), read while planning. Use its live models and exact
schemas, bind real reference assets, save receipts, recover outstanding jobs and stay within authorized scope.
Use the shared Draft/full-quality selection; honor an explicitly specified model. A prompt-only request ends with prompts, not media jobs.
The model generates any requested shot changes, speech, ambience, music or captions inside its output.
No Hypit, local stitching, external dubbing or promised later graphics repair is part of this skill.
Carry the chosen music treatment into the generation prompt and listening review. Captions follow the brief;
do not remove required speech or planned music after a generation error without resolving the mismatch.
Use [review and delivery](references/review-and-delivery.md) to check full playback, natural behavior, actual
words, lip sync, product fidelity, handling and identity. Fix the failed request within scope and inspect again.
Deliver the selected video or explicitly agreed clip set, complete prompts, task/settings/reference records
and accepted/rejected/unverified QA. Never equate successful submission with a convincing creator performance.
Referenced files: 21 — file references are metadata, not package contents.
