AI and Model Disclosure

AI Zombie Video is an independent zombie-themed AI visual-creation product operated by 温勇. It is not an official MiniMax, Doubao, Volcengine Ark or Cloudflare product and does not imply their endorsement.

Integrated processing paths

Provider and model	Role
MiniMax H3-Max, API identifier MiniMax-H3-Max	The current staged path uses two generated character images and the corresponding two uploaded-photo inputs as four references for two identities; historical paths may use two references
Volcengine Ark Seedream 5.0 Pro	Character-image generation for paths with an image-processing stage
Volcengine Ark Seedance 2.0	An alternative integrated video path selected through administrator configuration

This is the integrated model set, not a claim that every task calls all models. Each submitted task saves its configuration, model and price. Changes must be reflected in public disclosures; vague marketing names do not replace actual model identification. Verify this page against enabled model versions before public launch.

Inputs, outputs and limitations

Photos and necessary task settings are processed by the providers used for that task. Role exchange reassigns inputs; it is not authorization to replace one real person's face with another identity. Outputs may contain identity, anatomy, motion or scene errors; requests may be rejected or fail. Exact reproduction of references or examples is not guaranteed.

Concept illustrations express a direction and are not generation acceptance evidence. Real examples should identify their actual conditions. Label shared outputs as AI-generated and do not describe fictional scenes as real events. New videos may have audio removed by the delivery workflow; check the actual delivery specification.

Safety and personal rights

Use only authorized adult portraits and follow the Acceptable Use Policy. Permission does not replace a payment provider's eligibility decision about the actual generated scene.

Production processing includes platform-operated human review of reference images and composed instructions before generation, generated intermediate images before further processing, and the full video before delivery. Reviewers use an audited queue and maintained safety-term flags. This is human moderation, not an automated image/video safety classifier. Local development may operate without these gates; an updated worker and trained reviewers must be in place before public operation. Provider rejection, an upload checkbox and an AI label do not replace these reviews.

Data and charges

Processed inputs and delivery objects may have public links; originals have a separate private archive. See Privacy. Project credit prices are administrator-configured and are not provider tokens or a fixed currency conversion. Confirmed failed tasks return credits under the Refund and Credit Policy.