Seedance 2.0 Troubleshooting: Negative Prompts, Artifacts, and Fixes

When a generation fails, most people respond by adding more and more description. That often makes things worse. A better approach is to diagnose the failure mode and apply a targeted fix.

This post gives you a practical checklist and a set of "negative prompt" patterns you can append to your prompt (or rephrase as constraints) to reduce common issues.

A simple diagnostic loop

1. Keep the scene the same.
2. Change only one thing: camera, subject count, lighting, or motion.
3. Re-run and compare.

If you change everything at once, you cannot learn what caused the improvement.

Negative prompt block (general purpose)

Paste this at the end and adjust:

Avoid: extra limbs, duplicate faces, distorted hands, warped text, melted objects, jittery motion, sudden outfit changes, random camera cuts, heavy artifacts, over-sharpening, extreme blur.

If your workflow does not support explicit negatives, rewrite as constraints:

The video must have: stable anatomy, one face per character, clean hands, consistent wardrobe, smooth motion, no random cuts, no unreadable text.

Problem: faces drift or "become someone else"

Likely causes:

• too many characters
• unclear identity anchors
• extreme lighting changes

Fix prompt:

Single main character only. Keep the same face identity throughout.
Consistent hair and wardrobe: [repeat specifics].
Camera: stable, no rapid cuts, no extreme angle changes.

Problem: hands look wrong

Fix prompt:

Hands are visible and anatomically correct, five fingers per hand, no deformation.
Avoid: distorted hands, extra fingers, fused fingers.

Also: reduce fast hand motion. Replace "waving wildly" with "slow controlled gesture".

Problem: motion jitters or stutters

Fix prompt:

Smooth continuous motion, stable frame-to-frame consistency, no jitter.
Camera movement is slow and steady; no sudden accelerations.

Also: simplify the camera move. Pick one: dolly in OR orbit OR handheld.

Problem: the scene feels chaotic

Fix prompt:

One subject in the foreground, clean background with minimal movement.
No crowds, no random people entering frame.

Problem: text looks garbled

If you need readable text, keep it short and big:

On-screen text: "SEEDANCE" in large simple sans-serif, centered, high contrast, fully readable.
Avoid: distorted text, misspellings, unreadable letters.

If you do not need text, explicitly ban it:

No text, no logos, no watermarks, no subtitles.

Problem: lighting flickers or exposure shifts

Fix prompt:

Stable lighting with no flicker. Consistent exposure and white balance across the shot.

Problem: you asked for too much

This is the most common issue. If your prompt contains:

• 3+ locations
• 3+ major style references
• multiple time-of-day changes
• multiple camera moves

Cut it in half. Then cut it in half again.

A "clean cinematic" baseline template

When things break, return to baseline:

A single [subject] in [simple location], cinematic realism.
SHOT: medium shot, 35mm, shallow depth of field.
CAMERA: slow steady dolly-in, smooth motion.
LIGHT: soft key, controlled highlights, consistent color temperature.
Avoid: artifacts, jitter, deformations, random cuts, unreadable text.

If you want examples of prompts that already behave well, study the ones with video results in the collection: browse prompts.