Seedance 2.5 Guide

Prompt

Photorealistic nature documentary style with cinematic, true-to-life lighting. In the early-morning mist, across a vast wetland of reeds (scene references Image 1), an elegant red-crowned crane (appearance references Image 2) spreads its wings in a dance — feathers white as snow, a vivid red crown, a tall and slender frame. The setting is a shallow autumn wetland, the water glittering, surrounded by golden reeds, duckweed, and aquatic plants, with distant morning fog and the rising sun soft-blurred in the background. Low-angle medium shot with a subtle handheld feel, largely locked off, keeping the crane centered at all times. 0s-3s: The crane stands still in the shallow water and slowly spreads its broad black-and-white wings, the motion gentle, the airflow from its wingbeats rippling the surface. A morning breeze blows; sunlight slants in from the horizon, piercing the mist in a Tyndall effect. 3s-8s: The crane hops lightly across the water, its feet skimming the surface in turn and tossing up crystalline spray. Its long neck arches back, beak pointing at the sky; then it lands softly, folds its wings, sweeps its gaze gracefully around, and lets out a clear, ringing call. The low camera gently follows the crane's hops up and down, with natural depth of field and the foreground reeds softly blurred. Ambient nature sound only: flowing water, powerful wingbeats, and the crane's cry — serene, airy, and beautiful.

Requires 2 reference images (scene + subject); open with one sentence that sets the tone, break the plot down along a timeline in the middle, and close with the global mood — the four-layer structure reads at a glance.

Prompt

A 22-year-old East Asian woman with a gentle, classical face made for cinema. Cool-toned fair skin with a fine, supple texture, preserving the real fine pores and skin texture. Slender, peach-blossom eyes (the rims slightly moist), a softly relaxed brow, a small straight nose, full lips with a faint, barely-there tender smile, and a softly drawn jawline. Her gaze is deeply affectionate and full of feeling, a pool of spring water behind her eyes — tender, focused, carrying deep attachment and a trace of reluctance. Jet-black hair is coiled into an effortless, elegant low classical bun, held by a single plain jade hairpin, a few fine strands drifting loose along her cheeks and stirring in the breeze. She wears a minimalist, plain-white cross-collared robe of softly lustrous, understated silk, the collar slightly parted. Her frame is slender, her shoulders narrow, radiating a gentle, wistful, classically romantic air. Close-up framing, eyes looking straight into the lens, soft natural light, shallow depth of field.

Works as pure text-to-video; the same writing applies to animated characters too — just swap the "photoreal skin" language for the target art style.

Prompt

29-second one-take (a woman in period dress says goodbye by the river) [Global Scene Setup] Base environment: a riverside at dawn, a blurred small boat in the background — a lonely farewell scene. The mood is quiet and restrained. Visual style: cinematic, shallow depth of field (clean, blurred background), soft natural light. Camera language: hold a tight close-up on the woman throughout, as if from the subjective viewpoint of the man standing across from her. A subtle handheld breathing feel, no fast cuts. Character design: a 22-year-old East Asian woman with a gentle, classical face made for cinema. Cool-toned fair skin with a fine, supple texture, preserving the real fine pores and skin texture. Slender, peach-blossom eyes (the rims slightly moist), a small straight nose, full lips with a faint, barely-there tender smile. Her gaze is deeply affectionate and full of feeling, a pool of spring water behind her eyes. Jet-black hair coiled into a low classical bun, held by a single plain jade hairpin. She wears a minimalist, plain-white cross-collared robe of soft, lustrous silk. A slender frame radiating a gentle, classically romantic air. Performance core: restrained and fine-grained. Focus on the shifting gaze, the rise and fall of her breathing, the slight tremble of her lips, the tightening and release of her brow, the swallow in her throat, and the natural slide of tears. Prohibitions: no exaggerated sobbing, no fast cuts, no large body movements, no extra dialogue or BGM, and the tears must not fall early. [Emotion and Action Storyboard (building over 0-29 seconds)] Stage 1, 0-3s [Pressing]: she looks straight into the lens, no tears in her eyes yet, brow faintly knitted, lips parting as she murmurs softly: "Do you really have to go?" Subtext: she isn't venting — she's waiting for him to say out loud an answer she guessed long ago. Stage 2, 3-10s [Resignation]: her gaze slowly drifts away to the empty space beside her, eyelids lowering, the corners of her mouth pulling into a brief bitter smile that falls at once; her nostrils tighten slightly, she draws the smallest restrained breath, her chest rising once. Subtext: with this motion she is forcing the hurt and the bitterness back down. Stage 3, 11-17s [Remembering]: the camera pushes in slightly. She turns her eyes back and slowly, carefully scans his face. Her eye rims are red, but the tears are held back hard — not a drop falls. In the middle, a 0.5-second dead-silent pause: her lips stir once and press shut, her chin tightens, her throat rolls faintly. Subtext: this look is not goodbye — she is carving his face into memory one last time, for life. Stage 4, 18-23s [Lamenting]: she lowers her eyes, and without warning a single tear drops straight onto her collar. She doesn't wipe it; when she looks up again, the pleading in her gaze has turned into profound sorrow. The knitted brow begins to loosen, and she shakes her head — barely — once. Subtext: the relaxing brow signals inner release; the tiny head-shake is one last, helpless refusal of fate. Stage 5, 24-29s [Letting go]: the camera pushes to an extreme close-up. She works a soft smile onto her face (no self-mockery in it), and just as it reaches the corners of her mouth, a second tear slides from the corner of her eye past her nose and stops at her lip. In a near-inaudible voice she struggles to keep steady, she says: "Go." The smile stays frozen on her face as the tears keep sliding down, her gaze never leaving his. The camera finally rests on her face — smiling through tears, her eyes full of nothing but sorrow.

Global setup plus timestamped storyboard is the standard skeleton of a 30-second long take; timing prohibitions like "the tears must not fall early" are exactly the precision only 2.5 can hold.

Prompt

[Global Setup] Base environment and texture: an open-air roadside inn deep in a quiet bamboo forest at night. Push for top-tier 2D animation craft — the drift of falling bamboo leaves and the ripples in the wine must obey real physics, with the floating grace unique to Chinese-style wuxia. Visual style: premium donghua period look (cel-shaded 3D with realistic rendering), ink-wash edges. High-contrast interweaving of cold and warm light (desolate blue-white moonlight against the warm yellow candlefire on the inn's table). Camera language: razor-sharp montage cutting and multiple types of physically seamless transitions, with extreme contrast between stillness and motion. Character design: a 22-year-old swordsman with the handsome, sculpted face of donghua heroes. Sword-like brows and starry eyes, a high ponytail, loose strands over his forehead. He wears a white cross-collared combat outfit trimmed in black; a black-sheathed longsword rests on the table. Performance core: the killing speed of a top-tier wuxia "one-strike finish," the swordsman's unhurried composure, and the poetic negative space of Eastern aesthetics. Prohibitions: no greasy 3D look (strictly keep the pure cel-shaded donghua style). No sloppy fight choreography, no clipping or warped limbs. No subtitles and no extra dialogue. The ink-wash effect may only appear after the 25-second mark, with the "click" of the sheath — never earlier. [Timestamped Storyboard (with explicit cuts and transitions)] 00:00-00:04 Shot 1 [A Quiet Night's Drink] (match-cut morph): the camera holds a static low angle on the night sky, a huge, bright full moon dead center. The moon gradually dims and fades as a round wine bowl — identical in shape and size (top-down view of the table) — slowly materializes where the moon was and fully replaces it, moonlight dotted across the wine's surface with fine ripples. The transition must be seamless. The camera then slowly pulls back and down from the top view to eye level: the swordsman sits at the inn's table and slowly lifts the bowl. 00:05-00:10 Shot 2 [Hidden Killing Intent] (dolly-in close-up transition): the swordsman's motion hitches; his gaze turns suddenly razor-sharp. The camera drives forward at extreme speed, magnifying his eye until the pupil fills the screen — and the frame unfolds seamlessly inside the pupil: high above the bamboo forest, a black-clad assassin gripping a short blade reverse-handed dives out of the sky like a hawk, straight at the lens. 00:11-00:18 Shot 3 [Blades Cross] (action whip-pan transition): the camera yanks back to the swordsman's full-body framing as he leaps from his seat, drawing his sword. The camera follows his sword arm in a violent swing toward frame right (whip pan), smearing the whole screen into horizontal motion blur. Inside the blur the scene jumps to mid-air and freezes: the swordsman's longsword and the assassin's short blade collide hard, bursting into huge orange sparks and a ring-shaped shockwave where the weapons meet. 00:19-00:24 Shot 4 [One Strike Lands] (occlusion wipe): the two pass each other in mid-air. The shockwave of the exchange whips up a storm of bamboo leaves; one huge emerald leaf sweeps sideways across the lens, fully covering it and blacking out the frame. As the leaf drifts off, the picture has already cut to a ground-level eye-height shot: the assassin falls limply to the ground in the background while the swordsman, back to camera in the foreground, slowly sheathes his longsword. 00:25-00:30 Shot 5 [The Lingering Note] (ink-wash transition): with the "click" of the sword seating in its sheath, the entire bamboo forest and the fallen assassin suddenly dissolve like a drop of thick ink hitting water, spreading outward as black-and-white ink wash — and the frame freezes for good inside the splashed-ink landscape.

Each slice names one transition and spells out how it executes; negative timing constraints like "never earlier" are what keep the effect from jumping the gun.

Prompt

A joyful short set on a retro American cartoon-animation city street (scene references Image 5). The four young dancers' appearances, outfits, and facial features reference Image 1, Image 2, Image 3, and Image 4 respectively; the four dance in sync on the retro pedestrian street, their body movements tight and continuous, the spacing between them fixed. The cool-warm clash of the street neon stays unified throughout; locked-off frontal camera with a slow lateral tracking move. Every costume detail, facial feature, and body shape stays highly consistent with the reference images throughout — no face-swapping, no trading moves, each identity carries through, motion fluid with no clipping or distortion, in a retro cartoon-animation look.

Requires 5 reference images (4 characters + 1 scene); the official guidance is 1-8 subject images for best results — beyond that, stability drops and you'll be rerolling.

Prompt

On a rain-slick rooftop flashing with neon at night, the private detective from Image 1 leans against the railing and slowly raises his head toward the camera, rain dripping off his trench coat. In the voice of Audio 1 he says, low: "I found you." Keep Audio 1's low, suspenseful intonation, breathing rhythm, and gravelly texture, with lip movements strictly synced to the pronunciation. The only ambient sound is rain and distant traffic — no background music, no subtitles.

Requires 1 character reference image + 1 reference audio clip; "keep Audio 1's intonation and breathing rhythm" is the key phrasing for high-fidelity cloning.

Prompt

Reference the Shiba Inu's cute animated expressions, demeanor, movements, and creative pattern in Video 1 to make a short video of the black Shiba Inu from Image 1, precisely capturing the playful, lively state of the dog in the reference video. The setting is a bright living room with soft natural light, the camera gently following the dog's movements. Keep the black Shiba's coat color and physical features from Image 1 unchanged; the mood is warm and healing.

Requires 1 reference video + 1 subject image; "precisely capture the ... state/demeanor" triggers creative-level transfer far better than "do the same moves."

Prompt

Image 1 is a complete film storyboard containing 9 consecutive shots; each panel represents one full shot, together forming a complete, warm, healing cinematic narrative. The middle-aged woman references Image 2: a woman in her 40s-50s, 3D animated-film CG style, fair skin, short dark-brown curly hair, large eyes, a gentle, kind face, always with a natural smile. She wears a pale-yellow knit cardigan over a white inner layer and a blue ankle-length skirt — a homemaker-mother look; her appearance, wardrobe, and proportions stay consistent throughout. The kitten references Image 3: a round, chibi orange-and-white tabby kitten, 3D animated-film CG style, big eyes, a pink nose, fluffy soft fur. It wears a yellow knit hat and carries a green drawstring cloth bag — sweet and endearing; its coat, outfit, and proportions stay consistent throughout. Shot 1 (0-3s): evening in a city residential block; a medium shot at low angle slowly pushes in as the stray kitten hops onto a curbside trash can, flips the lid open with its front paws, and pokes its head in looking for food. Warm sunset light on the street; the mood is a little lonely but real. Shot 2 (3-5s): a close shot pushing slightly into the trash can; the kitten grips the rim with both paws and rummages inside, ears twitching, eyes focused — hunger and helplessness front and center. Shot 3 (5-9s): a locked-off medium shot at the apartment building entrance; the woman stands at the door holding food and spots the kitten. The kitten looks back at her, wary but not running — their first eye contact. Shot 4 (9-12s): a full-body medium shot; the kitten stands in the open yard and slowly approaches the doorway, hope in its eyes. Shot 5 (12-16s): the camera drops to the cat's eye level; the woman crouches and gently sets the food bowl in front of the kitten, which immediately lowers its head to eat, tail swaying lightly, while the woman watches with a smile — the frame full of warmth. Shot 6 (16-20s): a tracking medium shot; the woman walks slowly out of the yard carrying a basket of vegetables, the kitten quietly following beside her, always half a step behind. Shot 7 (20-23s): at the interior doorway, the woman slowly opens the door; warm yellow light spills out, forming a warm-cool contrast with the evening outside. Shot 8 (23-26s): a medium shot from behind and to the side; the woman turns back, smiles, and waves softly, inviting the kitten home. The kitten stands at the threshold looking up at her, ears perked, tail swaying, hesitating before gradually accepting the invitation. Shot 9 (26-30s): the camera slowly pulls back; the woman waits patiently at the door as the kitten finally steps forward and follows her inside, the warm yellow light stretching both their silhouettes long as the door slowly closes — the frame resting on the warm, quiet courtyard. Use a Pixar animated-film CG style, PBR physical materials, high-quality global illumination, soft volumetric light, cinematic depth of field. Camera movement natural and steady, the animals' behavior true to life, and continuity of time, space, character positions, and lighting maintained between shots. No face-swapping, distortion, or proportion errors on the characters; no extra unrelated people, animals, or props.

Requires 3 reference images (storyboard + woman + kitten); per-shot timestamps map one-to-one onto the storyboard grids for maximum fidelity.

Prompt

The overall style is a real commercial ad: cinematic photography, warm yellow lighting, a Tokyo-at-night atmosphere, natural and fluid visual language. Shot 1 (0-3s): the camera slowly pushes from a Tokyo street at night (streetscape references Image 1) toward the entrance of a ramen shop (storefront references Image 2). The shop sign, lanterns, business-hours board, and menu are all in natural, correct Japanese — no Chinese, English, or garbled text anywhere. Shot 2 (3-7s): the camera moves inside, where the clerk (appearance references Image 3) stands in the open kitchen, smiles at the camera, and says in natural, fluent Japanese: "Hello, would you like a bowl of our homemade tonkotsu ramen?" His mouth shapes stay strictly synced to the Japanese pronunciation, the voice natural, with the characteristics of a real Japanese male speaker. The background keeps the real ambient sounds of boiling noodles, steam, and clinking tableware. Shot 3 (7-11s): cut to a close-up of the ramen (references Image 4), steam rising continuously, the chashu swaying slightly, the oil on the broth flowing naturally. Brand caption text appears on screen, all in natural, correct Japanese: "Rich flavor, a bowl of happiness," in a typeface true to Japanese commercial-ad design. Shot 4 (11-15s): the customer (appearance references Image 5) sits at the table, tastes the ramen, breaks into a satisfied smile, and says naturally to the camera: "It's really delicious — I'll definitely be back!" The voice keeps natural Japanese pronunciation with accurate lip-sync. The brand logo (references Image 6) and a Japanese slogan appear: "A bowl I crave every day." All signage, menus, posters, and logos stay in unified Japanese — no other language mixed in; every line of dialogue in the video is in Japanese, and no Chinese subtitles or dubbing may be generated. Character identities stay consistent — the clerk and the customer never swap faces or outfits.

Requires 6 reference images (street / storefront / clerk / ramen / customer / logo); the constraint "everything in the target language, no mixed-in text" must be spelled out hard.

Prompt

A vertical 9:16 e-commerce outfit showcase video against a pure-white, minimalist seamless studio backdrop (layout references Image 1), the fixed split layout holding throughout: on the left side of the frame, three rectangular modules with thin black rounded borders stacked vertically, each topped by a black serif numeral (1, 2, 3), each module containing a flat-lay of the corresponding outfit piece with a small "1x" label beneath it; on the right side, a young female model stands full-body displaying the outfit for real — full-body eye-level locked-off camera, soft even studio lighting, no hard shadows, fabric textures crisp and fine. The model holds a natural, elegant stance with a few slow motions — a hand rising to adjust the clothing, a shift of posture — the overall style clean and premium. Absolutely no text captions, slogans, watermarks, logos, or extra annotations at any point; no background music, no voiceover, no ambient noise — a completely silent, purely visual picture; the camera stays locked with no push, pull, pan, or tilt, and the layout structure stays unchanged throughout.

Requires 1 layout reference image, and the audio toggle is already switched off for you at generation time; the "no subtitles + no BGM + silent" trio finally executes reliably on 2.5.

Prompt

Remove all the background music from Video 1, keeping only the voices of the people speaking and the ambient noise. The vocals must keep their original timbre, tone, breathing, and detail; the picture, people, subtitles, image quality, color grade, and composition all stay completely unchanged — do not alter any visual content.

Requires 1 reference video with vocals; this is the go-to move for fan edits, interview cleanup, and swapping in multi-language dubs.

Prompt

Remove the woman on the left and the man on the right from Video 1, keeping only the woman in the center, her movements unchanged. Fill the removed areas naturally to match the original footage's lighting and perspective, leaving no ghosting or blurry patches; every other part of the picture, the composition, and the camera movement stay exactly the same.

Requires 1 source video; name who stays and who goes by position, add one line about clean fill-in — that's the highest-success recipe for invisible removal.

Prompt

Completely remove the green English title text that appears in Video 1, and edit the blue sportswear worn by the person doing back stretches into red sportswear — change only the clothing color. Also fully remove the extra fitness props in the background: the blue kettlebell on the right and the blue balance pad on the left. The person themself, their skin, head, hands and feet, body posture, exercise movements, the yoga mat, the carpet, the wall, and the camera frame all stay unchanged. After the text and props are removed, the areas they covered must be filled in naturally — no text ghosting, no blurry patches, no prop remnants. The red sportswear must stay stable as the person rolls over, lifts a leg, and stretches, keeping the original folds and fabric texture of the clothing.

Requires 1 source video; multiple edit targets can go in one prompt, but each target needs its own clear "change into what."

Prompt

Reference the camera movement and physical lighting changes of Video 1, and precisely replace the white model in the video with the elf queen from Image 1. Her appearance strictly follows the image: long silver-white hair, a magnificent white-and-gold embroidered robe, and a magic staff topped with a glowing crystal — keep the character consistent, no clipping, no face changes. The setting is an elven great hall, keeping the scene from Video 1. The camera positions match the original video, and the elf queen gracefully raises her magic staff following the model's movements in the reference video. As she moves, her silver-white hair and the wide skirt of her white robe flow naturally with the air currents, silky with real drape. Along with the motion, the lighting effects from the reference video transform into a sacred pillar of light rising from the ground, echoing the dazzling golden glow of the staff in her hand. The final video requires cinematic CG photoreal quality, sacred and epic lighting, a high-definition picture, finely detailed skin, crisp metallic textures on the costume, rich detail, and fluid, natural motion.

Requires 1 white-model video + 1 character reference image; "precisely replace the white model with the character from Image N" is the core phrasing of white-model rendering.

Prompt

Turn the entire white background of the video into a pure green-screen background. The person's movements, the foreground elements, and the frame composition stay unchanged; the edges of the person are clean with no leftover white fringe, and hair detail stays intact. Remove the audio and keep it silent.

Requires 1 source video, and the audio toggle is already switched off for you at generation time; once you have the green-screen plate, swap the background in any editing software.

Prompt

Seamlessly join Video 1 and Video 2 without modifying either one. The carp in Video 1's final frame starts swimming rapidly forward, then leaps quickly out of the water into the air, where a heavy, ancient stone gate appears; the carp crosses over the gate into the clouds and mist, finally transforms into a dragon, and the shot connects to Video 2.

Requires 2 videos (the opening and closing clips); the line "without modifying the source videos" is mandatory, and the transition description covers only the missing middle stretch.