Seedance prompts: 18 examples and a practical guide

How do you write a Seedance 2.5 prompt?

For Seedance prompts, name the subject and reference image, one main action, one camera move, the light and ambient sound. This guide applies that structure to Seedance 2.5 in SpicyFrame. Controls and supported inputs can differ between model versions and interfaces; do not assume that a workflow documented for Seedance 2.0 has identical settings here. The 18 examples below are editable starting points, not measured benchmark results. Start with one image and a short scene, then review the complete output before spending on longer versions.

Seedance 2.5 prompt guide: the six-part structure

Write each prompt as subject + action + camera + light + mood + audio. Subject says who or what is on screen and which file to use: “@Image 1, wearing the outfit from @Image 2”. Action is one visible movement with a speed: “turns slowly toward the camera”. Camera is one move or a locked frame: slow push-in, side tracking shot, gentle orbit. Light gives direction and colour: “warm golden-hour backlight” or “soft window light from the left”. Mood is two or three words: relaxed, confident, cinematic. Audio describes the ambient sound to generate: rooftop wind, soft rain, café murmur. Finish with what must not change: “keep face and outfit unchanged”.

Using @Image references correctly

In SpicyFrame you can add one to 12 JPG, PNG or WEBP references under 10 MB each and refer to them as @Image 1, @Image 2 and so on, in upload order. Give every reference a job: “@Image 1 for face and hair, @Image 2 for the dress, @Image 3 for the terrace.” A reference that is uploaded but never mentioned is ambiguous, and two references that disagree, such as different haircuts, push the model to average them. Start with one sharp lead photo, add others only when the shot needs information the lead image lacks, and keep the subject, lighting family and styling consistent across the set.

How long should a Seedance 2.5 prompt be?

Aim for roughly 25 to 60 words. Shorter prompts leave the camera and light to chance; much longer prompts usually contain contradictions, such as a locked camera and an orbit, or three actions competing for five seconds. Use plain verbs and concrete nouns rather than stacked style words: “slow push-in” is more useful than “epic ultra-cinematic masterpiece camera”. The studio quality selector controls the output: Medium 480p, High HD 720p or Ultra High Full HD 1080p. Writing a resolution in the prompt does not change the purchased setting. When a render misses, change one element at a time so you know which instruction made the difference.

Matching the prompt to 5, 10 or 30 seconds

Five seconds fits one gesture: a blink and smile, a turn, lifting sunglasses. Ten seconds fits an action that settles: two steps and a pose, a slow walk into light. Seedance 2.5 can render 30 seconds in a single pass, which suits a continuous scene with a clear arc, so describe an opening, a middle and an ending: “begins looking away, turns toward the camera midway, ends with a slow smile as the camera stops.” Test any new prompt at five seconds in Medium 480p for 399 piments ($3.99), then order the final length in High 720p, where 10 seconds costs 899 piments ($8.99) and 30 seconds 2,499 ($24.99).

Portrait and close-up prompts

1. “@Image 1 blinks, then smiles softly at the camera, soft window light from the left, very slow push-in, calm and intimate mood, quiet room tone, keep facial features unchanged.” 2. “@Image 1 turns the head slightly toward a warm lamp, locked camera, amber light on the cheek, relaxed evening mood, faint city ambience.” 3. “@Image 1 looks up from a book and meets the camera, shallow depth of field, morning daylight, gentle push-in, soft page rustle.” Close framing keeps the face large, which makes identity easier to hold and to check.

Fashion and glamour prompts

4. “@Image 1 walks toward the camera on a runway in the outfit from @Image 2, flashbulbs, low-angle tracking shot, confident mood, crowd murmur and camera clicks.” 5. “@Image 1 turns slowly under red and violet neon, haze in the air, slow push-in, sultry and cinematic, low electronic hum.” 6. “@Image 1 in a satin evening dress steps onto a hotel balcony, city lights below, side tracking shot, glamorous mood, distant traffic.” 7. “@Image 1 adjusts a tailored blazer and looks over the shoulder, clean studio light, gentle orbit of a few degrees, editorial mood, quiet studio ambience.”

Outdoor, travel and lifestyle prompts

8. “@Image 1 walks along a rooftop at golden hour, hair moving in the breeze, side tracking shot, warm backlight, relaxed mood, wind and distant city.” 9. “@Image 1 lifts sunglasses and smiles by the pool, sparkling reflections, slow push-in, bright summer mood, gentle water lapping.” 10. “@Image 1 walks through a rainy street at night, neon reflections on wet asphalt, steady tracking shot, moody atmosphere, rain and footsteps.” 11. “@Image 1 sips coffee at a café terrace and smiles, fixed medium shot, soft morning light, easy mood, café chatter.” 12. “@Image 1 stands on a beach as waves roll in, dress moving in the wind, slow pull-back, sunset glow, peaceful mood, surf.”

Product, mood and 30-second prompts

13. “@Image 1 stays centered while a soft highlight sweeps across the bottle, slow push-in, dark studio, premium mood, quiet hum, preserve the label exactly.” 14. “Candle flames flicker beside @Image 1, who turns slightly toward the light, slow dolly-in, warm intimate mood, soft restaurant ambience.” 15. “@Image 1 stands by a rain-streaked window, subtle breathing, locked camera, cool blue light, contemplative mood, rain on glass.” 16, for 30 seconds: “@Image 1 begins at the far end of a neon corridor, walks slowly toward the camera, pauses under a pink light at the midpoint, then looks over the shoulder as the camera settles, steady tracking shot, cinematic night mood, low synth ambience.” 17, for 30 seconds: “@Image 1 sits on a balcony at dusk watching the city, then turns and smiles at the camera as the lights come on, slow push-in throughout, warm-to-blue light shift, soft wind.” 18, prompt only: “A model walks along a quiet harbour at sunrise, side tracking shot, pastel light, calm mood, gulls and water.”

Common Seedance 2.5 prompt mistakes

The most frequent mistakes are several actions in five seconds, two camera moves that conflict, references uploaded but never named, asking for readable text, logos or signs, and requesting angles the photos never show, such as a full back view from a front portrait. Vague mood words without a light direction and quality tags instead of instructions also waste renders. Do not expect exact dialogue or a specific song: SpicyFrame generates ambient audio and does not accept uploaded music or scripts. Avoid explicit or anatomical wording too; SpicyFrame does not produce explicit content and the provider can refuse such prompts. Rewrite toward wardrobe, setting, light and camera instead.

Build a prompt without starting from a blank page

The free Seedance prompt builder lets you assemble an action, setting and camera direction in your browser. Preparing the text does not generate a video and does not require a purchase. Check every part against your reference before using it: a walking instruction is a poor match for a tightly cropped face. For more scenarios, use the image-to-video prompt library; for output options, read the Seedance image-to-video guide.

A practical review loop for your first render

Save your source image, prompt, selected duration and quality before ordering. After generation, inspect the opening, middle and final frames, then watch the complete clip with sound. Compare visible identity, hands, clothing, background edges and camera motion with the brief. If the action is too large, reduce it; if the camera wanders, request a locked shot. Keep the other instructions unchanged so the next result is easier to assess. Every new render is a separate paid request. A clearer prompt improves the brief but does not guarantee that the model will follow every detail.

Seedance 2.5 prompt structure at a glance

AI image-to-video questions answered

Sources

External pages, opened in a new tab. Rankings reflect the time of writing.

1. Runway: Image-to-Video Prompting Guide
2. Adobe Firefly: writing effective prompts for video generation
3. ByteDance research: Seedance 2.0 model report

Related guides