Why Seedance 2.0 Rejects Reference Video — and What Omni Reference Actually Does

You found a video that's exactly the style you want, uploaded it as a reference, and Seedance 2.0 rejected it. The error message doesn't explain why. You try again with a different video — same result.

Before assuming the platform is broken, it's worth knowing that "reference video" means two entirely different things in Seedance 2.0 depending on which mode you're using. Most rejections happen because the video is going into the wrong input slot, not because the video itself is invalid.

TL;DR

• Seedance 2.0 has three distinct reference input modes. Mixing them causes silent failures.
• Omni reference is the multi-modal mode: it accepts up to 9 images + 3 videos + 3 audio files as style guides. Reference videos here guide motion and atmosphere — they don't get animated directly.
• First-frame / last-frame mode animates a static image as the starting frame. Videos don't go here; images do.
• The most common rejection reasons: wrong mode, real human faces in the video, copyright content, unsupported format.
• Fix for face-related rejections on seedance2.so: use the Portrait model tier, which handles real-person content directly.

---

The three reference modes in Seedance 2.0

This is the root cause of most "reference video rejected" confusion. The three modes are distinct and not interchangeable:

Mode 1: Text-to-video

No reference input. Write a prompt, the model generates entirely from scratch. The model decides composition, characters, style, and motion.

Mode 2: First-frame (image-to-video)

A single image becomes the literal first frame of the video. The model animates it forward. This mode accepts images only — not videos. Uploading a video in this slot will fail.

You can also supply a last-frame image to constrain both the opening and closing frame, with the model generating the motion in between. Both first-frame and last-frame inputs are images, never videos.

Mode 3: Omni reference

This is where reference videos belong. Omni reference accepts up to 9 images + 3 reference videos + 3 audio files as creative guides. The model reads these references for motion style, atmosphere, character appearance, and audio timing — then generates new video content that follows those cues. The reference videos are not animated; they inform the generation.

In your text prompt, you tag the references by index: @image1, @video1, @audio1. The model reads these tags and applies the referenced material as guidance for the corresponding aspect of the output.

These modes are mutually exclusive. You cannot combine first-frame inputs with omni-reference inputs in the same generation. Attempting to do so is a common source of silent rejections.

---

Why Seedance 2.0 rejects reference videos

If you're in the right mode and still getting rejections, there are four causes to check in order:

1. Real human faces in the video

The same face-detection rule that applies to reference images also applies to reference videos in omni-reference mode. If the video contains identifiable real human faces, the standard model blocks it before generation begins.

Fix: On seedance2.so, switch to a Portrait model tier (tagged "Portrait" in the model selector). Portrait-tier models are face-enabled and accept real-person video content directly.

Alternatively, if you're only using the video for motion style rather than character appearance, you can substitute an image reference and describe the motion in your text prompt instead.

2. Copyright content in the video

Footage containing branded logos, recognizable franchise visuals, or copyright-flagged material gets rejected even without explicit named references. The detection layer runs on visual content, not just your prompt text.

Fix: Use original footage or material you have rights to. If you're trying to reference a motion style from a commercial film or branded content, describe the motion in your prompt text instead of uploading the clip directly.

3. Mixing input modes

Trying to supply both a first-frame image and a reference video in the same generation will fail. The two modes don't coexist.

Fix: Decide which mode you actually need:

• Want a specific character to appear and animate → first-frame mode with an image of the character
• Want to guide the motion, style, or atmosphere of a generated scene → omni-reference mode with the reference video

4. Format or duration issues

Reference videos need to be in a supported format (MP4 recommended), within a reasonable duration, and publicly accessible if you're supplying a URL. Very long clips or unusual codecs may be silently rejected.

Fix: Trim the reference clip to 5–30 seconds and re-encode as MP4 before uploading. For URL references, confirm the URL is publicly accessible without authentication.

---

What omni reference is actually for

Omni reference is Seedance 2.0's multi-modal generation mode. The name comes from the range of input types it accepts — images, videos, and audio — as creative references all at once.

You assemble a set of references that collectively describe what you want in the output:

• @image1 might be a character portrait — the model uses it for character appearance
• @video1 might be footage of someone walking on a beach — the model borrows the motion pattern, not the person
• @audio1 might be a music track — the model syncs the visual pacing to the beat

None of these inputs get directly animated. They're references, not source material. The model generates entirely new video content that reads those references as a creative brief.

This is what makes omni reference powerful for workflow use cases: you can maintain consistent characters across multiple generations by using the same image reference each time, vary the scene while keeping the character stable, or lock in a motion style from a reference clip while changing everything else.

The limit is up to 9 image references, 3 video references, and 3 audio references in a single generation. You can use any combination within those limits.

---

Omni reference vs. first-frame: which to use

---

Where to try it

The reference-to-video studio on seedance2.so supports omni-reference mode with multi-image and multi-video input. For generations involving real human faces, select a "Portrait" tagged model — these are face-enabled and won't reject footage with real people in it.

Free credits on signup. No credit card required.