Consented Romantic Video

Mission

Create a non-explicit romantic video draft—such as an embrace or kiss—from authorized adult reference images or an approved starting frame. Preserve the distinction between a creative depiction and a real event. This workflow never authorizes non-consensual face merging, sexualized deepfakes, impersonation, or the use of minors or public figures.

Mandatory consent gate

Require confirmation that every identifiable adult has authorized:

use of their image or likeness;
the described romantic action;
the intended audience and distribution; and
any generated audio, voice, or promotional association.

Treat a supplied photo, public profile, relationship claim, or “they would be okay with it” as insufficient. If consent is unknown, stop and request an approved fictional or adult source. Do not proceed with minors, coercion, non-consensual intimacy, public-figure likeness, or a scene intended to deceive someone about a real relationship or event.

Required inputs

one or more authorized adult reference images or an approved parent frame;
participant IDs, reference roles, and what identity details must remain;
non-explicit scene, setting, wardrobe, action, camera, and duration;
target ratio, resolution, audio, disclosure, and intended use; and
reviewer, variant count, budget, and rights/consent record.

Do not assume gender, relationship, age, orientation, or appearance from a photo. Do not automatically stitch two portraits or use face-swap tooling when a reference-capable image-to-video path can preserve roles more transparently.

Choose the generation path

Use MODELS.md and the live schema:

one approved composite/start frame: I2V with a hosted image_url;
multiple authorized references: a model-specific images_list or image_urls path with explicit participant order; or
a supplied romantic parent clip: use the editing workflow instead of regenerating identity.

Veo, Wan, Gemini Omni, Kling, and other families accept different image-list fields, durations, audio defaults, and ratios. Resolve the exact endpoint through muapi-video-tools.md. Do not copy fields from the source application's model list.

Execute and review

Record consent state and freeze the scene description. Keep the action affectionate and non-explicit.
Upload references and record role/order, provenance, returned URLs, and retention expectation without storing secrets.
Render one proof before variants. Retain prediction ID and poll the result.
Review identity, age presentation, facial boundaries, hands, contact, wardrobe, temporal stability, unwanted nudity, audio, framing, and synthetic disclosure. Use video QA.
Return parent/child relationships, failed or blocked branches, and the smallest safe iteration. Do not describe the output as evidence of a real relationship or event.

Output contract

Return participant IDs, consent/distribution state, reference roles, scene brief, model/endpoint, prediction ID, output URL, status, disclosure label, QA findings, and any blocked requests. Save permitted receipts under .video/ using reports-and-receipts.md.

Generation is draft-only. Publishing or sharing requires the participants' explicit approval and any applicable synthetic-media disclosure.