Dreamina Prompts: How to Write for Seedream and Seedance

Dreamina (dreamina.capcut.com, from ByteDance and CapCut) generates images with Seedream and video with Seedance, and the two want different prompts. Most frustration with the tool comes from writing one kind of prompt for the other: a scene description pasted into the video tool, or a motion-heavy sentence given to the image tool. This guide sets out a structure for each, shows worked examples, and lists the habits that make prompts fail.

A structure for Seedream image prompts

Seedream responds well to natural sentences rather than keyword lists. Write them in this order so the model reads the most important thing first:

1. Subject with one or two defining details. "A teenage skateboarder in an oversized orange hoodie."
2. Setting. "On an empty concrete plaza after rain."
3. Composition. "Low angle, wide shot, subject centred, lots of sky."
4. Light and mood. "Overcast, soft reflections in puddles, muted palette."
5. Medium or finish, if needed. "Photographic, 35 mm look" or "flat illustration, thick outlines."

Each sentence does one job. If you find yourself writing "and" three times in a line, split it. Specific nouns beat adjectives: "oversized orange hoodie" gives the model more to work with than "cool outfit".

Leave out quality boosters such as "ultra detailed" or "masterpiece". Recent models ignore them at best and, at worst, push the image towards a generic glossy look. State the finish you want instead.

Motion prompts for Seedance video

When you animate a still, the image has already answered "what does this look like?". The prompt should answer "what happens?":

• One subject action. "The skateboarder pushes off and rolls towards the camera."
• One camera behaviour. "Camera holds still" or "slow tracking shot alongside".
• Pace words. "Slowly", "gently", "subtle" keep motion within what a short clip can hold.
• What stays fixed. "Background and lighting remain constant."

Avoid re-describing the clothing, the plaza or the weather. Repeating the scene invites the model to re-render it, which is where colour shifts and flickering come from. If you are generating video from text alone, combine both structures: scene sentences first, then one motion sentence.

The Dreamina image and video guide covers the settings side of the same workflow.

Worked examples

Product shot. "A matte black ceramic mug on a pale linen cloth. Studio setting, single softbox from the upper left, gentle shadow to the right. Close-up, slightly above eye level, mug off-centre. Clean, minimal, neutral background." Then for video: "Steam rises slowly from the mug. Camera holds still."

Character illustration. "A young woman with short silver hair and round glasses, wearing a green raincoat. Standing under a tram shelter at night. Medium shot, eye level. Neon reflections on wet pavement, cool blues with one warm sign. Flat illustration, limited palette." For video: "She looks up as rain intensifies. Subtle camera push-in."

Landscape. "A single pine on a rocky ridge above a fog-filled valley. Dawn, pink light catching the treetop, fog in deep blue shadow. Wide shot, horizon in the lower third. Photographic." For video: "Fog drifts slowly through the valley. Camera holds still."

Settings at a glance

Control names in the current interface may differ a little from these, and credit costs vary by model and quality, so check the platform’s current pricing page.

Why prompts misfire

• Too many subjects. Three characters doing three things rarely resolves cleanly. Pick one focal action and let the rest be background.
• Negatives written as positives. "No text" often adds text. Use a negative prompt field if one is available, or simply avoid mentioning the thing.
• Contradictory lighting. "Golden hour" and "overcast" in one prompt produce a muddle. Choose one.
• Brand names and real people. Both are unreliable and raise rights questions. Describe the object or the look instead.
• Lettering in the image. Keep any text short, in quotation marks, and expect to retry. For video, add titles in an editor afterwards.

Try it

Image (Seedream): A vintage bicycle leaning against a sunlit yellow wall, a paper bag of oranges in the basket. Narrow street, late afternoon, long shadows. Medium shot, eye level, bicycle slightly left of centre. Photographic, warm and natural.
Video (Seedance): The shadows lengthen slowly and a leaf drifts past. Camera holds still. Colours stay consistent.

Change one sentence at a time. Swap the time of day first, then the composition, and keep notes on which wording produced the version you liked.