Plan draft video-ad variations from supplied performance evidence, preserving the distinction between observed data and creative hypotheses.
中文简介正在准备中。当前显示原文说明。
Build a consented continuity system for an authorized person, fictional character, product, or world across video shots.
中文简介正在准备中。当前显示原文说明。
查看安装方法安装前可以查看交给 AI 代理的指令内容。
Reduce identity, wardrobe, prop, geography, and chronology drift across a multi-shot sequence. Treat the continuity sheet and approved reference set as first-class inputs. This skill supports authorized fictional characters, consenting performers, and owned products; it is not a license to imitate a real person.
Before creating or editing a likeness, confirm the user's authorization for the person, voice, image, or performance and the intended use. Do not proceed with non-consensual impersonation, sexualized deepfakes, deceptive political material, or a minor's likeness. For a public figure, require a clearly authorized production context and label synthetic output where appropriate.
Keep the following states separate: unknown, user-supplied, consent claimed, consent verified, and approved for publication. A source URL or
local file alone is not proof of permission.
For each entity, record:
Use one reference role per field. Do not send a product pack as a character reference or reverse the order of a first/last-frame list. Keep reference URLs and local IDs in an asset manifest rather than burying them in prose.
Use MODELS.md and a live schema check. Reference-capable families such as Seedance omni-reference, Wan reference-to-video, Gemini Omni character/reference flows, and Kling Omni may be useful, but their accepted reference counts, placeholder syntax, audio defaults, and duration controls differ. Use only fields the selected endpoint documents.
If an image reference is missing, hand off to the adjacent image agent or use an explicitly available image capability. Do not generate a new identity anchor without making the substitution visible to the user.
Return the continuity bible, asset-role map, authorization state, model/schema
choice, parent/child graph, output URLs and job IDs, identity-drift findings,
failed shots, and recommended next iteration. When permitted, save the bible,
manifest, receipt, and QA note under .video/ with the formats in
reports-and-receipts.md.
Never claim a likeness is exact or a product is faithful without a review. Generation remains draft-only and publication requires separate approval.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Plan draft video-ad variations from supplied performance evidence, preserving the distinction between observed data and creative hypotheses.
中文简介正在准备中。当前显示原文说明。
Create authorized presenter, avatar, and UGC-style video drafts with explicit voice, likeness, consent, and synthetic-media handling.
中文简介正在准备中。当前显示原文说明。
Analyze a script or existing edit, identify visual gaps, and generate supplemental B-roll coverage with clear parent and factual-status labels.
中文简介正在准备中。当前显示原文说明。
Translate creative intent into coherent cinematography, framing, camera, lens, lighting, motion, and sound direction for a video shot.
中文简介正在准备中。当前显示原文说明。
Create a non-explicit romantic scene from authorized adult references with clear consent, identity boundaries, and synthetic-media labeling.
中文简介正在准备中。当前显示原文说明。
Turn one approved topic into a curiosity-led 3D educational short with character sheets, ordered keyframes, motion clips, narration, captions, and an auditable assembly plan.
中文简介正在准备中。当前显示原文说明。