Seedance 2.5: Complete Prompting Guide (Full Prompt Library)

A tested prompt library for Seedance 2.5: 10 categories with the exact settings and reasoning behind each one, plus how to structure a prompt that holds up.

Most Seedance 2.5 prompts only ever get run once, so there's no real way to know if they'd hold up a second time. This is a library of 10 strong prompt categories: drama, action, commercial, horror, and more. Each category was tested across multiple generations to see which prompt structures held up consistently. This guide explains what Seedance 2.5 can do, how to structure a prompt for it, and what we learned from the tests.

What Is Seedance 2.5

Seedance 2.5 is ByteDance’s video generation model for clips up to 30 seconds, with native audio, up to 50 multimodal references and region editing. The model generates natively up to 1080p.

Up to 50 reference images can go into a single generation too, locking a face, an outfit, a location, or a color palette in place before a single frame renders. And if one detail still comes out wrong, region edit fixes just that spot without forcing a full re-render of the clip.

Spec

Detail

Native resolution

Up to 1080p

Max clip length

Up to 30 seconds

Aspect ratio

Any ratio from vertical 9:16 to widescreen 21:9

Audio

Ambience, foley, and score generated in the same pass as the video

Character consistency

Holds identity, wardrobe, and lighting across shots from a single reference

How to Write a Strong Prompt for Seedance 2.5

A prompt works best as one continuous block of text, broken into labeled sections. Skip a section and the output tends to fail in a specific, predictable way.

GLOBAL STYLE comes first: genre, color grade, film stock or digital look, aspect ratio, shutter behavior, and anything that shouldn't appear. Everything after this section needs to match it.

SCENE is a one-line logline, what's happening, where, in what mood. Save the detail for later sections.

CHARACTERS describes every person in the shot: face, hair, build, wardrobe. If a reference image already covers this, just note which reference is which character.

LOCATION covers the space and its props, separate from the people in it. A vague location is the most common reason a multi-shot sequence starts drifting between cuts.

FIRST FRAME AND BLOCKING pins down exact starting positions, who's where, facing which direction, right as the clip begins. This gives the model something fixed before any motion starts.

The shot-by-shot breakdown comes next, each one labeled "Shot 1", "Shot 2", and so on, with the shot type and action in a sentence or two, ending in "Hard cut" wherever a cut belongs. Pacing gets built here, not in a separate field.

OPTICS and CAMERA describe the lens and camera movement for each shot: focal length, camera height, handheld or dolly or crane.

PHYSICS covers how fabric, smoke, hair, and liquid should move, anything that would look off moving like a solid object.

LIGHTING goes further than GLOBAL STYLE on the light source: where it's motivated from, its direction, how it falls on faces or surfaces.

AUDIO comes last: ambient noise, specific sound effects, and what shouldn't be there, usually "No music, no discernible dialogue" unless the scene needs otherwise.

Most working Seedance 2.5 prompts follow the same shape: one visual rule at the top, one sound rule at the bottom, and everything in between broken into shots.

10 Prompts in Practice

We tested each category a few times each to see if the same setup keeps working, or if it only looks good the first try. Below, for every category: the model, the settings we used, the actual prompt, and the video it produced.

Dramatic Exterior: Human Subject, Available Light

Settings: Genre: Drama. Lighting: Golden hour, single motivated source. Emotional control: Grief held beneath composure. Camera: Anamorphic large-format, mixed wide and close coverage.

Why this combination: Drama holds shots longer than most genres, so the two-person scene has time to play out. Golden hour keeps the light the same across every segment without describing the sun's position in each one. The wide shots in Segments 1 and 5 hold the ocean and the couple together in one frame. The closer lenses in Segments 2 and 3 stay on each face for the two lines, with the background soft and out of the way.

Prompt used:

Action Sequence: Multi-Shot Choreography

Settings: Genre: Action. Physics: Realistic, high impact, real paper mass and inertia. Camera: Anamorphic, hybrid handheld and gyro-rig. Lighting: Hard cold key with a single warm accent.

Why this combination: Action tightens pacing and keeps the camera close to the movement, so a nine-shot fight reads as one continuous sequence instead of separate takes stitched together. Cold lighting with a single warm accent keeps the eye on the heroine and the cash without extra description needed for every cut. Real paper physics on the money, individual bills catching light and drifting on their own air currents, keeps hundreds of loose banknotes from reading as one rigid mass.

Prompt used:

Commercial Product: Multi-Reference Scene

Settings: Genre: General, everyday realism as the baseline for contrast. Camera: True handheld with light shake, walking follow and orbit shots. Physics: Real magic on honest rules, one clean light-transfer moment, then radial color waves.

Why this combination: Locking three references, the girl, the location, and the product, keeps all three stable while the actual gimmick, a color wave that freezes every bystander, happens around them. Light handheld shake sells the scene as something filmed in a real park, since the color effect is already the most stylized part of it. Building the recolor as a fixed rule, waves spreading from each footfall at constant speed, gives the model one mechanic to repeat the same way every time instead of improvising it shot to shot.

Prompt used:

Epic Landscape: Environment as Protagonist

Settings: Genre: Epic. Camera: Slow, steady push-in across all three shots, ground-level, low angle, and aerial. Lighting: Bright diffused daylight, cold cinematic grade.

Why this combination: Epic spreads the composition wide and makes the environment itself the subject, which works when there's no person in the shot to hold interest instead. The same slow push-in on all three shots builds one sense of scale instead of three separate ones. Cold diffused daylight keeps the glacier, rock, and grass reading as real physical space, not a stylized backdrop.

Prompt used:

Noir Scene: Shadow and Atmosphere

Settings: Genre: Drama, noir base. Lighting: Hard key, high contrast, monochrome. Camera: Anamorphic, locked lens per segment, five-segment sequence. Color: Black and white throughout, zero color anywhere.

Why this combination: A hard overhead key with deep shadow builds the noir look directly, one narrow source carving the hat brim shadow across the eyes instead of an even scene graded black and white afterward. The lens lock per segment, wide at 84° for the establishing shots down to 18° telephoto for the two-shot, sets how much the background compresses without spelling out depth of field in the action lines. The same screen direction on every cut, the sedan always from the right, the woman always moving right to left, keeps the street readable across all five segments.

Prompt used:

Multi-Character Scene: Consistent Faces Across Frames

Settings: Genre: Action, high-energy performance piece. Camera: Relentless dynamic camera, hard cuts and match cuts on the beat, average shot under 1.5 seconds. Lighting: Grounded cinematic contrast per location, no neon anywhere. Dual quality system: polished [FILM] as default, degraded [REPORTER CAM] for crowd and helicopter beats only.

Why this combination: Four named characters through twenty-four fast cuts is a hard test of face consistency, since there's no time to re-establish who's who between cuts. The member-count lock, exactly four, never five, never duplicated, stops the crowd shots from blurring an extra into the cast. Switching cleanly between clean digital cinema and degraded phone or helicopter footage keeps the two looks from bleeding together, so it's always clear which four faces are on screen.

Prompt used:

Fantasy Action: Character and Creature at Scale

Settings: Genre: Epic, action. Physics: Hyperbolic speed, no blood anywhere. Camera: Anamorphic, handheld weight with operator correction, wide and telephoto mix. Emotional control: Calm, focused, not afraid.

Why this combination: Hyperbolic scale, a horse flying at roughly 65 meters per second against a five-meter archdemon, makes the creature read as genuinely massive instead of a big model standing still. Locking the sword to cuts only, edge first, never a stab, gives the model one combat grammar to repeat across all five sword actions instead of a different swing type each time. Replacing blood with glowing light, sparks, and ash keeps every demon death consistent and avoids the gore that would break the rest of the world's rules.

Prompt used:

UGC-Style Ad: Natural Movement, No Production Feel

Settings: Genre: General, no genre-specific visual logic imposed. Camera: Handheld selfie framing, fixed for the whole video. Lighting: Soft even cabin light, natural daylight through the window. Emotional control: Understated, self-conscious, no performing.

Why this combination: UGC works because it doesn't look produced, so a fixed selfie framing held for all thirty seconds, with real turbulence in the opening shots, sells authenticity better than any lighting setup would. Understated acting, natural blinks, reactions arriving a half-beat late, small smiles, is what makes this read as a real reaction instead of a scripted ad. Locking the seating geometry, the same aisle seat and background passengers across all eight shots, keeps the video from drifting into something that looks staged.

Prompt used:

Horror Scene: Wrong Light, Wrong Angle

Settings: Genre: Horror. Camera: Deep one-point-perspective, 4:3 aspect ratio. Physics: Real weight and inertia. Emotional control: Tired, welling tears, held tension without release.

Why this combination: Horror pulls light and framing away from where a viewer expects them, and deep one-point-perspective aisles with the girl small against empty corridors turn the negative space past her shoulder into the actual threat. Keeping the antagonist out of focus until the final second, always a blur behind shelves, builds dread from absence instead of a direct reveal. Real weight on ordinary objects, sneakers squeaking, boxes toppling, keeps the setting grounded, so the wrongness reads as a genuine intrusion instead of a stylized effect.

Prompt used:

Documentary Style: Observational Camera

Settings: Genre: General, third-person observational. Camera: Single fixed 47° lens for the entire sequence, distance-only framing changes. Lighting: Five distinct natural states across one continuous span of time, high sun to blue hour.

Why this combination: A documentary feel needs the camera to look like it's observing, not directing, so one fixed focal length for all thirty seconds, with framing changed only by walking closer or further, keeps the sequence from reading as constructed. Crowd size doubles as the clock, forty people at the start growing to a hundred at the peak and thinning to five or six by dawn, showing one continuous night instead of five separate scenes. Rules against the camera lingering on the body, no slow pans, no isolated shots of torso or legs, keep the tone honest instead of staged.

Prompt used:

What the Testing Shows

Reference images hold identity across cuts more reliably than a text description does, once a face or a product is locked as a reference, it survives hard cuts, lighting changes, and camera moves that would otherwise cause drift.

Genre and lighting settings do more work than most of the prompt text itself. Changing genre alone shifts pacing, contrast, and camera behavior even when the written scene description stays identical.

Multi-character scenes are the hardest category to keep consistent. The more faces in a single shot, the more reference material each one needs to avoid drifting into a generic face by the third or fourth cut.

The shot-by-shot breakdown format outperforms a single continuous scene description for anything with more than one camera angle. Vague transitions between shots are the most common cause of a sequence falling apart.

50 references is a ceiling worth using deliberately, not maxing out by default. A cluttered reference set with too many competing faces, props, or locations produces a less coherent result than a smaller, carefully chosen set.

The GLOBAL STYLE and AUDIO sections function as guardrails more than instructions. Explicitly excluding what shouldn't appear, modern artifacts, music, dialogue, prevents more failures than adding more positive description does.

Seedance 2.5: Complete Prompting Guide (Full Prompt Library)

Got any questions left?

by TAVOA AI

Discover more

Best 5 Platforms to Access Seedance 2.5 in 2026

Seedance 2.5 on TAVOA AI in 2026: What You Get and How It Works

Seedance 2.5 Pricing: Full Breakdown of Credits, Plans, and Cost per Clip

11-Day Seedance 4K Unlimited: August 2026 Promotion