What you will make
A Seedance clip that goes through, built on a Seedream 5 picture, with dialogue in a language the model speaks well.
You need
- 1.The exact error message you got
- 2.The prompt and the files you used
- 3.An idea for the character or product in the shot
1. Find your error
The message tells you what failed: a picture or file, the prompt, the soundtrack or a setting. Each one has its own fix.
Reference may show a real person, or "input image may contain real person". A picture you attached looks like a photo of a real person, and realistic AI faces trip it too. Fix: make the picture again in Seedream 5, step 2.
Tip: Change one thing per retry, so you know which change fixed it.
2. Make the picture in Seedream 5
Seedream and Seedance come from the same model family, so Seedance accepts a realistic person made in Seedream. The same person made with other models often gets blocked as a possible real person.
- 1In the Media Studio, open the Model card, Image tab, and pick Seedream 5.0 Pro.
- 2Describe the face, the outfit and the place. To fix an existing picture, attach it and ask for the same person.
- 3Open the result and press Use Image.
PromptA candid photo of a woman in her early thirties behind the counter of a small corner coffee shop in the morning. Olive skin, dark curly hair tied up with a few loose strands, light freckles, a small silver hoop in one ear, calm focused eyes. She wears a rust orange apron over a cream linen shirt with the sleeves rolled up. She is pouring steamed milk into a cup on the counter. Behind her: a chalkboard menu with no readable words, shelves of jars and ceramic cups, a hanging plant, warm light from a window on the left. Shot at eye level, natural morning light, soft shadows, realistic skin texture, shallow depth of field. No text, no logos, no lettering, no watermarks anywhere in the frame.
Tip: Only for people you made up. A photo of a real person, a celebrity or a lookalike stays blocked, and it should.
3. Put the picture in Seedance
The Seedream picture goes in as a reference. The prompt says what moves and what you hear, not what the picture already shows.
- 1Model card, Video tab, Seedance 2.5 or Seedance 2, References tab.
- 2Add the Seedream picture.
- 3Write the camera, the action and the sound.
Prompt@image1 is the barista. Keep her exact face, hair and outfit in every shot. She finishes pouring the milk, slides the cup across the counter toward the camera and looks up with a small smile. Slow push in from a medium shot to a close up. Handheld, realistic, morning light. Sound: the hiss of the steam wand, a cup set down on wood, quiet cafe chatter. No subtitles, no text, no watermark, no logo.
Tip: To animate one picture on Seedance 2, use the References tab. The Frames tab on Seedance 2 needs two pictures, a first and a last.
4. Describe the music, never name it
Seedance makes its own soundtrack. When it sounds like a known song, the whole clip is stopped.
- 1Take out song titles, artist names and real lyrics.
- 2Describe the genre, the mood, the tempo and the instruments.
- 3Keep Include Audio on.
PromptMusic: slow, warm lo-fi hip hop with soft brushed drums, a mellow electric piano and a round bass line, quiet under the room sound.
Tip: Turning the sound off does not fix it. Describing the music does.
5. Pick a language it speaks well
Seedance 2.5 speaks 11 languages: Chinese, English, Spanish, Portuguese, Japanese, Korean, Indonesian, Malay, Thai, Vietnamese and Arabic. Seedance 2 is safest in Chinese and English.
- 1Write each spoken line in the language you want to hear.
- 2Name the language before each line: In English, warm and calm: "Here you go."
- 3Keep each line in one language. Names and brands are fine.
Tip: Romanian is on neither list. Make the clip without dialogue and add the voice in Text to Dialogue.
6. Stop subtitles, logos and face drift
These go wrong without an error, so the prompt has to prevent them.
- 1End the prompt with: No subtitles, no text, no watermark, no logo.
- 2Use one clear, front facing picture per person.
- 3Pick a landscape ratio when you can. Portrait clips get burned in subtitles more often.
Tip: Generate shows the cost before you press it. Fix the prompt first, then render.
Avoid these
- Using a realistic person made in another image model instead of Seedream 5.
- Naming a song or an artist in the prompt.
- Turning Include Audio off to get past the copyright block.
- Using the Frames tab on Seedance 2 with only one picture.
- Mixing two languages in one spoken line.
- Changing the picture, the prompt and the settings all at once.
Seedance 2 and Seedance 2.5 errors
Seedance checks every picture you attach for anything that looks like a real person, and it cannot tell a real photo from a realistic AI face. Make the person in Seedream 5 instead: it is the same model family as Seedance, so Seedance accepts it.
Seedance makes its own soundtrack, and when it sounds like a known song the whole clip is stopped. Take song titles, artist names and lyrics out of the prompt, describe the genre, mood, tempo and instruments instead, and keep Include Audio on.
Eleven, for the prompt and for spoken lines: Chinese, English, Spanish, Portuguese, Japanese, Korean, Indonesian, Malay, Thai, Vietnamese and Arabic. Name the language before each line that is not Chinese.
Chinese and English. Seedance 2 has no published language list, and its official prompt guide covers those two. For spoken lines in any other language on the Seedance 2.5 list, use Seedance 2.5.
You are in the Frames tab. On Seedance 2 it makes a transition between two pictures, so it needs both. To animate one picture, add it in the References tab instead. Seedance 2.5 Frames works from one first frame.
End the prompt with "No subtitles, no text, no watermark, no logo", and use pictures without text in them. Landscape clips get burned in subtitles less often than portrait ones.
No. Seedance does not accept photos of real people, and the Seedream 5 route is for people you made up. Use an original character, or one whose face is not shown.
Models used
Open a guide to see what each model is best at.
Your turn
Everything in this lesson is one click away. Open the tool and make your first version now.

