Seedance real-face reference images rejected: what to do

BytePlus ModelArk states that Seedance 2.5 and Seedance 2.0 series models do not support directly uploading reference images or videos that contain real human faces. If a face-containing reference is blocked, do not try to disguise it. Use references that have no real face (illustrated, 3D, or generated characters, objects, scenes), or use the exceptions BytePlus describes for its own platform.

The rule is from the BytePlus page, read 2026-10-01; this is not legal advice. Sume facts are from its Video generation docs.

What are the BytePlus exceptions?

Two are described. First, trusted model outputs: face-containing outputs generated on ModelArk by certain models, under the same account, within the past 30 days, can be used as input assets again. The page lists conditions: only ModelArk outputs, only the same account, only original outputs (not after secondary editing or after the validity period), and compressing or forwarding files may invalidate trust. Second, authorized real-person assets that went through the real-person verification and onboarding process.

What does this mean on Sume?

Sume's docs describe reference inputs as guidance: reference images are used "rather than exact frames", and Seedance 2.x models honor audio and video references. The sources I read do not describe the BytePlus trust mechanism on Sume, so do not assume it carries over. The seedance-2.5 row on Sume covers 480p, 720p, and 1080p at 4 to 30 seconds.

What is a safe reference plan?

Build the character without a real person's face: generate or illustrate it first, then pass that image as the reference. Keep the likeness question separate; for a person's photo see private figure likeness without consent. If a request is blocked, read the error rather than retrying the same input, and check the per-model limits in Seedance 2.5 references on BytePlus vs Sume.