Seedance 2.5 is just around the corner, and we had the chance to test the model ahead of its wider availability.
Naturally, we wanted to see what it could actually do.
We gave Seedance 2.5 six very different 30-second prompts: two fantasy fights, two grounded action sequences, a five-member K-pop performance, and a quiet countryside journey. Together, they test camera direction, character consistency, choreography, scene geography, pacing, audio, and prompt adherence.
Below are the results, our early impressions, and every prompt we used. Copy them, modify them, and try them with your own ideas when Seedance 2.5 becomes available on Yapper.
This is an early hands-on test. Capabilities and availability may change before the model's wider release.
Seedance 2.5 is ByteDance's next-generation audio-video model. It is built around longer, more controllable generations rather than isolated short clips.
The official Dreamina overview highlights four major capabilities:
- Up to 30 seconds of native video generation
- Up to 50 multimodal references, including images, video, audio, scripts, and style material
- More precise reference and motion control
- Localized video editing for targeted fixes without regenerating an entire clip
The 30-second limit is the most important change in practice. It gives a prompt enough time for setup, escalation, and payoff inside one generation. Instead of stitching together several short clips and hoping the character, lighting, and location match, you can direct a more complete scene in one pass.
Our early results were promising:
- Long prompts can hold a real sequence. The model followed complex beat sheets across a full 30 seconds instead of collapsing everything into one repeated action.
- Visual identity is generally strong. Faces, costumes, lighting, and locations stayed surprisingly stable, especially in the simpler two-character and single-character tests.
- Important continuity notes can survive several cuts. In the first fantasy test, the damaged tree remained visible after the sword impact, exactly as requested.
- Prompt adherence is not shot-for-shot. Dense action was sometimes compressed or restaged, and the K-pop test occasionally showed six performers despite a five-member prompt.
- Complex choreography is still the hardest problem. Fast fights, prop interactions, exact group counts, and precise spatial relationships are more likely to drift than mood, lighting, or overall story direction.
The model is already good at translating production language into convincing images. The next step is making every count, contact point, and movement land with the same reliability.
The first prompt used a full three-act structure: a quiet forest setup, a sword fight with a persistent tree-impact detail, and a final creature reveal.
A 30-second forest confrontation with a persistent tree-impact detail and a final creature reveal.
This was one of the strongest overall results. The forest atmosphere, character identities, camera coverage, and escalating action all remained coherent. Most impressively, the sword's impact left a visible gash in the tree and that damage stayed present in later shots. The final creature reveal also arrived as a clear payoff instead of an unrelated cutaway.
Create a 30-second cinematic fantasy confrontation between a rugged human ranger and an elegant dark elf warrior inside an ancient mist-covered forest.
0–4 seconds — Establishing wide shot: A vast primordial forest at dawn, towering trees disappearing into fog, ancient stone ruins covered in moss. The ranger moves cautiously through the environment while the dark elf watches silently from a high branch.
4–8 seconds — Medium tracking shot: Follow beside the ranger as he draws a weathered longsword and scans the forest. His cloak moves naturally in the wind. Leaves and mist drift through shafts of warm sunlight.
8–11 seconds — Extreme close-ups: The ranger's eyes react to a faint sound. Cut to the dark elf's hand slowly drawing a narrow silver blade. Close-up of both weapons catching the light.
11–15 seconds — Dynamic wide shot: The dark elf leaps from the tree and lands in front of the ranger. The camera performs a fast circular move as both fighters immediately enter combat.
15–19 seconds — Medium combat shot: The elf attacks with rapid, precise strikes. The ranger blocks twice, steps backward, and swings a powerful horizontal counterattack.
19–22 seconds — Tree-impact hero shot: The dark elf narrowly ducks beneath the ranger's sword. The ranger's blade slams violently into the trunk of a large ancient tree. Bark explodes outward in realistic wooden fragments and dust, leaving a deep fresh gash in the trunk. The sword remains embedded for a brief moment.
22–25 seconds — Close-up and recovery: Tight close-up of cracked bark, falling splinters, and the ranger pulling the sword free with visible effort. The elf uses the opening to move behind him.
25–28 seconds — Slow-motion confrontation: The ranger turns and blocks the elf's blade inches from his face. Sparks scatter while loose bark continues falling in the background.
28–30 seconds — Epic wide reveal: Both fighters suddenly stop and turn toward an enormous ancient creature awakening behind the ruins. The damaged tree trunk remains clearly visible in the foreground. End as two glowing eyes open through the fog.
Camera: Sweeping crane shots, medium tracking coverage, aggressive low angles, controlled handheld combat movement, fast circular camera motion, macro inserts, and wide IMAX-scale reveals.
Visual style: Photorealistic dark fantasy, cinema realism, ARRI Alexa 65 color science, IMAX large-format composition, anamorphic depth, subtle organic film grain, cinematic motion blur, realistic bark destruction, and detailed skin, fabric, and metal textures.
Lighting: Cool forest shadows, warm sunrise beams, volumetric fog, soft magical highlights, and realistic exposure roll-off.
Audio: Forest ambience, distant birds, boots on leaves, sharp blade impacts, a heavy sword strike against wood, cracking bark, falling splinters, a deep creature rumble, and a rising orchestral score.
Continuity constraints: Maintain identical ranger and elf faces, costumes, body proportions, weapons, and hairstyles throughout. Keep the tree damage visible after the impact. The sword must strike the tree only once. Preserve clear spatial geography and readable combat choreography.
Avoid: Rubber-like tree movement, an exploding entire tree, excessive debris, the sword passing through the trunk, changing weapon design, extra limbs, floating objects, cartoon effects, gore, text, logos, or watermarks.
The second fantasy prompt removed the quiet setup and began in the middle of the fight. It asked for continuous escalation, faster cuts, and a decisive ending.
A faster fantasy combat test built around continuous escalation, visible tree damage, and a decisive ending.
Seedance 2.5 handled the change in pacing well. The clip feels materially faster than the first test, while still preserving the two fighters and the forest location. The tree impact, the ranger's loss of control, and the final defeat all read clearly. Some choreography was simplified, but the overall progression remained intact.
Create a 30-second relentless cinematic fantasy fight between a rugged human ranger and an agile dark elf warrior inside an ancient mist-covered forest. Begin in the middle of the battle with no introduction or establishing pause.
0–3 seconds — Immediate action opening: Start on a tight medium shot as the ranger blocks a rapid dark elf sword strike. Their blades collide beside the camera, throwing sparks across the frame. The ranger is already breathing heavily and backing through the forest.
3–6 seconds — Dynamic wide tracking shot: The dark elf presses forward with a fast combination of precise attacks. The ranger parries while moving between massive tree trunks, roots, and ruined stone structures. The camera tracks alongside them at high speed.
6–9 seconds — Close combat inserts: Tight close-up of steel scraping against steel, the ranger's boots sliding through wet leaves, the elf rotating around his guard, and the ranger barely blocking another strike.
9–13 seconds — Aggressive counterattack: In a medium handheld shot, the ranger shoves the elf backward and swings his longsword in a powerful horizontal arc. The elf ducks at the final instant.
13–16 seconds — Tree-impact shot: The ranger's sword crashes violently into the trunk of a huge ancient tree. Bark, splinters, and wood dust burst outward. The blade cuts a deep fresh gash into the trunk and becomes embedded.
16–19 seconds — Immediate punishment: With no pause, the dark elf kicks the ranger in the ribs while he struggles to free the sword. The ranger is thrown against the damaged tree. The elf follows with another rapid strike as the ranger pulls the sword free just in time.
19–23 seconds — Escalating forest brawl: Fast alternating wide, medium, and close-up shots of the ranger blocking, stumbling, and countering. They crash through hanging branches and ruined stonework. The elf remains faster, constantly attacking from changing angles.
23–26 seconds — Hero wounded: The dark elf slips inside the ranger's defence and cuts across his upper arm or side. Keep the injury non-graphic. The ranger drops to one knee, loses his grip, and the sword falls into the leaves.
26–29 seconds — Final attack: The ranger tries to rise, but the elf strikes him with the sword hilt and drives him backward. He hits the damaged tree trunk and collapses heavily to the forest floor.
29–30 seconds — Final close shot: Hold on the ranger lying injured among wet leaves, struggling to remain conscious. The dark elf steps into frame and points the blade toward him. Hard cut to black.
Editing: Very fast, snappy cinematic cuts; no slow motion; no pauses; no calm opening; continuous escalation from the first frame to the ending.
Camera: Aggressive lateral tracking, controlled handheld combat coverage, whip pans, low-angle pursuit shots, tight weapon inserts, wide shots preserving fight geography, and a sharp final close-up.
Visual style: Photorealistic dark fantasy, cinema realism, ARRI Alexa 65 color science, IMAX large-format scale, anamorphic depth, fine organic film grain, realistic cinematic motion blur, detailed bark destruction, natural skin texture, and weathered metal and fabric.
Lighting: Cold forest shadows, sharp beams of dawn light through mist, moving highlights across blades, realistic exposure roll-off, and atmospheric depth.
Audio: Immediate sword impacts, boots crushing leaves, strained breathing, metal scraping, a heavy strike against wood, cracking bark, falling splinters, body impacts, and intense orchestral percussion.
Continuity constraints: Maintain identical ranger and elf faces, hairstyles, costumes, body proportions, and weapons throughout. Preserve the tree's visible damage after impact. The ranger's injury must remain visible and consistent. The sword strikes the tree only once.
Avoid: Slow motion, giant creatures, magical interruptions, excessive visual effects, gore, rubber-like movement, exploding trees, the sword passing through the trunk, random weapon changes, extra limbs, duplicated characters, floating objects, text, logos, or watermarks.
Next, we moved from fantasy to a grounded action-thriller scene with an armed entrance, a gunfight, a transition to hand-to-hand combat, and a final fall through a window.
A grounded action test that moves from an armed entrance and gunfight into a hand-to-hand brawl.
This test demanded more props, more characters, and a larger amount of physical destruction. The model kept the office layout and the two leads readable through the escalation, while the desk, papers, gunfire, filing cabinets, and window all contributed to the same scene. The precise shot list was compressed in places, but the gunfight-to-fistfight transition worked.
Create a 30-second intense cinematic action sequence set inside a criminal warehouse office. Begin inside the room with no exterior establishing shot. The action should feel immediate, grounded, and physically realistic.
0–3 seconds — Wide office reveal: Inside a dim drug-lord office overlooking a large warehouse floor. The drug lord sits behind a heavy wooden desk surrounded by cash, papers, and industrial monitors. Two armed guards stand nearby. Harsh fluorescent lights flicker above.
3–6 seconds — Explosive entrance: The office door suddenly bursts inward. The hero charges through the doorway holding an MP5-style submachine gun. Wood fragments and dust scatter through the room as everyone reacts.
6–10 seconds — Fast gunfight montage: Medium tracking shot as the hero fires controlled bursts while moving sideways into the room. Close-up of muzzle flashes lighting the hero's face. Papers, glass, and office objects erupt around the drug lord's desk. The guards dive for cover.
10–13 seconds — Table flip: The drug lord violently flips the heavy desk onto its side and drops behind it. Cash, documents, and a desk lamp fly through the air. Bullets strike the overturned tabletop, tearing away wood splinters.
13–17 seconds — Intense exchange: Low-angle shot from behind the overturned desk as the drug lord returns fire. Cut to a wide shot showing the hero moving between concrete pillars and office furniture. Sparks burst from metal surfaces and windows crack behind them.
17–19 seconds — Empty magazine: Tight close-up of the hero pulling the trigger and hearing an empty click. Insert shot of the depleted magazine. The hero lowers the weapon as the drug lord realizes the advantage.
19–22 seconds — Transition to hand-to-hand combat: The drug lord rushes forward. The hero throws the empty weapon aside and blocks the first strike. They collide violently against a filing cabinet.
22–26 seconds — Brutal office brawl: Medium handheld shots of punches, grappling, and body impacts. The drug lord slams the hero into the wall. The hero counters with an elbow, drives the drug lord backward across the office, and knocks furniture aside. Keep the choreography readable and weighty.
26–28 seconds — Window struggle: Both men crash into the large warehouse-facing window. The glass fractures across the entire frame. They continue grappling at the edge while broken glass begins falling outward.
28–30 seconds — Final static wide shot: Hold on the office window from inside the room. The hero forcefully pushes the drug lord backward, but the drug lord grabs him. Both men smash through the window together and disappear from view. Glass fragments fall through the empty frame. Do not show the landing. End on the shattered window.
Editing: Rapid, snappy cuts during the gunfight, slightly longer handheld shots during the fistfight, constant escalation, no pauses, and no slow introduction.
Camera: Wide room geography, medium tracking shots, weapon and reaction close-ups, low-angle desk coverage, controlled handheld brawl cinematography, and a static final composition.
Visual style: Grounded action-thriller cinema realism, ARRI Alexa 35 color science, IMAX-scale composition, subtle organic film grain, realistic skin, wood, metal, and glass textures, cinematic motion blur, and strong highlight roll-off.
Lighting: Cold fluorescent warehouse lighting, warm desk lamps, intermittent muzzle-flash illumination, dusty volumetric atmosphere, and reflections across the office windows.
Audio: Door explosion, sharp gunfire echoes, shell casings, wood impacts, glass cracking, an empty magazine click, heavy breathing, punches, furniture breaking, and a sudden music drop during the final fall.
Continuity constraints: Maintain identical hero and drug-lord faces, clothing, body proportions, and injuries throughout. Keep the same office layout and window position. The hero's magazine must be empty before the fistfight. Both characters must fall through the window together.
Avoid: Graphic wounds, visible gore, slow-motion bullet shots, unlimited ammunition, random weapon changes, excessive muzzle flashes, rubber-like body movement, characters passing through furniture, glass exploding before impact, showing the bodies landing, text, logos, or watermarks.
The fourth test simplified the cast but made the setting more specific. We asked for a weathered office in northern Mexico, a green-and-amber neo-noir palette, and only two people in the entire sequence.
A two-character office gunfight and brawl with a locked visual palette and a final fall through the window.
This version benefited from the tighter constraint. The room, light, wardrobe, and two-character geography stayed clear from the door burst to the empty final window. The visual direction also carried through: stained walls, harsh sunlight, sickly fluorescent light, dust, and damaged furniture all belonged to the same world.
Create a relentless 30-second cinematic action sequence inside a rugged drug lord's office in an old concrete building in northern Mexico. The location feels authentic, weathered, humid, and lived-in rather than luxurious: cracked plaster walls, faded green paint, exposed wiring, rusted metal shutters, a dusty ceiling fan, cheap religious candles, old photographs, stained tile floors, and a heavy wooden desk. Sickly green fluorescent light mixes with harsh amber sunlight leaking through the shutters, creating a gritty green-tinted neo-noir atmosphere.
There are only two characters: the hero and the drug lord. No guards, assistants, or background characters.
Begin instantly as the wooden office door bursts inward. Use snappy cuts: the lock breaking, the drug lord turning in surprise, the door slamming against the wall, and the hero charging through dust and splinters with an MP5-style submachine gun.
The hero immediately fires while advancing. Rapidly alternate between wide room geography, medium lateral tracking, close-ups of the hero's focused eyes, the MP5 firing, shell casings bouncing across dirty tiles, and bullets destroying bottles, framed photographs, lamps, and plaster walls. Most shots last between half a second and two seconds.
The drug lord violently flips the heavy desk onto its side and dives behind it. Papers, cash, cigarette smoke, and desk objects scatter through the room. Bullets tear into the desk, producing realistic wood splinters and dust. The drug lord returns fire with a handgun from behind cover.
Use aggressive shot-reverse-shot editing, whip pans, match-on-action cuts, and rapid inserts of bullet impacts, sparks, cracked tiles, shattered glass, and green fluorescent tubes flickering overhead. Maintain clear screen direction and readable positioning.
The hero moves between a concrete pillar, an old metal cabinet, and damaged furniture. The ceiling fan continues rotating slowly above the chaos. Dust and smoke hang in the green light.
The MP5 magazine runs empty. Show three sharp cuts: the hero pulling the trigger, the distinct empty click, and his immediate reaction. He throws the weapon aside just as the drug lord rushes him before he can reload.
Transition instantly into a brutal fistfight. The hero blocks the first punch, and both men crash into a metal cabinet. Use snappy close-ups of fists, elbows, boots, hands gripping shirts, strained faces, and bloodless impact details, mixed with short handheld medium and wide shots that keep the choreography clear.
They exchange heavy punches, grapples, knees, and body slams through the destroyed office. Furniture breaks, the desk shifts, candles fall, and papers swirl around them. Cut on every major hit, block, throw, and collision.
The drug lord slams the hero into a cracked wall. The hero counters with an elbow, drives him backward, and forces him toward a large open window with old wooden shutters overlooking a narrow courtyard below.
Both men collide with the window frame while grappling. The remaining glass cracks but does not immediately break. Use fast inserts of spreading fractures, boots sliding across dusty tiles, hands gripping clothing, and the shutters violently swinging.
The hero pushes the drug lord backward, but the drug lord grabs him and pulls him along.
End with one locked-off wide shot from inside the ruined office. Both men smash through the window together and disappear from view. Glass, broken shutters, cash, and papers fall through the empty frame. Do not show the landing. Hold briefly on the damaged office, rotating ceiling fan, drifting smoke, and empty window, then hard cut to black.
Editing: Extremely snappy hard cuts, whip pans, match-on-action cuts, impact frames, rhythmic cutting synchronized with gunfire and punches, no slow motion, no dissolves, no dead air, and no long takes before the ending.
Visual style: Gritty neo-noir action realism, green-tinted color palette, ARRI Alexa 35 color science, IMAX composition, subtle organic film grain, natural skin texture, weathered architecture, practical debris, deep shadows, dusty highlights, and cinematic motion blur.
Audio: Splintering door, sharp indoor gunfire, shell casings on tile, bullets striking wood and concrete, buzzing fluorescent lights, rotating ceiling fan, an empty-magazine click, heavy breathing, punches, breaking furniture, cracking glass, and intense percussion synchronized with the edits.
Continuity: Preserve identical faces, clothing, weapons, room layout, lighting, and visible damage. The MP5 must run empty before the fistfight. Both men must fall through the window together.
Avoid: Bodyguards, crowds, luxury interiors, cultural caricatures, gore, unlimited ammunition, random weapons, excessive camera shake, rubber-like movement, premature window destruction, showing the landing, unreadable action, text, logos, and watermarks.
The K-pop test changed almost every variable at once: five performers, synchronized movement, rapid formation changes, individual close-ups, a huge stage, moving lights, and beat-matched editing.
A large-scale performance test with synchronized movement, formation changes, close-ups, stage lighting, and confetti.
The final clip looks polished and convincingly staged. Wide shots, close-ups, the runway, lighting, and confetti all feel like parts of the same award-show performance. This was also the clearest example of a remaining limitation: the prompt asked for five performers, but several shots contain six. Visual style and energy were easier to preserve than an exact group count.
Create a bright, energetic 30-second K-pop choreography performance on a massive award-show stage. Feature a polished five-member idol group performing sharp, perfectly synchronized choreography with confident expressions, powerful formations, and precise musical timing.
Begin immediately with the group already dancing at the center of the stage. Open on a dramatic ultra-wide shot revealing a large LED backdrop, illuminated runway, moving spotlights, audience light sticks, and sweeping stage lasers. Cut sharply into a medium frontal shot as all five performers hit the same powerful pose on the beat.
The choreography should feel fast, athletic, and highly coordinated: crisp arm movements, synchronized footwork, sharp body isolations, rapid formation changes, controlled jumps, and strong finishing poses. Keep every performer clearly visible during major group movements.
Use snappy, rhythm-driven editing with cuts synchronized to the music. Mix ultra-wide stage shots, medium group coverage, close-ups of individual performers, low-angle hero shots, overhead formation shots, side-tracking camera moves, and fast crane sweeps. Most shots should last between half a second and two seconds.
Show the group transitioning from a straight-line formation into a V formation, then rotating positions while maintaining perfect synchronization. Include a brief center-member highlight where the camera pushes forward as the lead performer delivers a sharp dance break, while the other members continue coordinated movement behind them.
Build toward a high-energy chorus section. Use a sweeping wide shot as the group moves down the illuminated runway, followed by quick close-ups of confident expressions, synchronized hand gestures, boots striking the stage, hair moving naturally, and stage lights flashing on the beat.
Near the ending, the performers execute a rapid sequence of perfectly matched movements, followed by a synchronized jump and immediate formation lock. Finish with a powerful final pose at center stage as gold confetti cannons fire, the LED wall erupts into bright abstract graphics, and the audience lights fill the arena. End on a low-angle hero shot, then hard cut to black.
Editing: Extremely snappy hard cuts, match-on-action cuts, whip-pan transitions, rhythmic close-up inserts, no slow motion, no dead air, no long static shots, and constant visual energy.
Camera: Dynamic ultra-wide, wide, medium, close-up, and extreme close-up coverage; fast dolly movement; crane sweeps; overhead formation shots; low-angle hero framing; and smooth tracking around the group.
Visual style: Premium K-pop award-show realism, ARRI Alexa 35 color science, IMAX-scale composition, subtle organic film grain, clean skin tones, glossy stage reflections, cinematic motion blur, vibrant highlights, rich contrast, and polished broadcast-level production.
Lighting: Bright cyan, pink, gold, and white stage lighting; synchronized strobes; moving spotlights; LED floor reflections; controlled lasers; dramatic backlighting; and no overexposed faces.
Audio: Upbeat K-pop dance track with punchy drums, bright synths, crowd cheers, stage impacts, and choreography hits synchronized tightly to the beat.
Continuity: Maintain identical faces, hairstyles, outfits, body proportions, and performer positions throughout. Preserve clear group formations and realistic choreography spacing.
Avoid: Duplicated performers, extra limbs, warped hands, changing outfits, unstable faces, random camera shake, unreadable formations, excessive motion blur, floating objects, illegible text, logos, and watermarks.
For the final test, we deliberately removed the rapid edits, fights, debris, and ensemble choreography. The prompt asked for one traveller, one bicycle, natural movement, and patient golden-hour cinematography.
A quiet lifestyle test following one traveller from a misty morning walk to a lakeside sunset.
This was the cleanest and most consistent result of the group. The traveller, wardrobe, bicycle, fields, lake, and changing light all fit together naturally. The slower pacing gave the model more room to preserve identity and compose attractive shots. It is a useful reminder that a model can appear much more reliable when the scene is designed around actions it can render clearly.
These tests also revealed a practical prompt structure for longer AI video generations.
A 30-second prompt works better when it has explicit beats. You do not need to direct every frame, but the model should know what must happen first, what changes in the middle, and what the final image should be.
Separate Story, Camera, Style, and Audio
Do not bury every instruction inside one paragraph. Give the model separate direction for:
- The visible action and timing
- Camera distance and movement
- Editing rhythm
- Lighting and visual style
- Sound and music
- Continuity rules
- Known failure modes to avoid
If a damaged tree, empty magazine, injury, wardrobe detail, or character count must remain visible, say so in the relevant beat and repeat it in the continuity section. Those reminders help the model distinguish essential story information from optional decoration.
The best prompts in this test had one clear technical challenge: the tree strike, the empty magazine, the window fall, the formation lock, or the bicycle-at-sunset ending. Build the sequence around that moment and give it enough screen time to read.
Seedance 2.5 looks like a meaningful step forward, especially for creators who want to generate complete scenes rather than isolated shots.
Across these six tests, the model showed strong cinematic composition, attractive lighting, good character continuity, and enough prompt adherence to carry a setup, escalation, and ending across 30 seconds. The best results felt less like disconnected AI clips and more like directed sequences.
It is not perfect. Exact counts can drift, dense choreography can be simplified, and props or spatial relationships can change across fast cuts. Those limitations matter, especially when a scene depends on precise action rather than atmosphere.
Still, the longer generation window changes what is possible. It gives creators more room to build a story, establish a location, develop an action, and land a final image in one generation.
Seedance 2.0 also improved substantially after launch. Based on that track record and these early tests, we are optimistic that Seedance 2.5 will become an even stronger creative option over time.
Create with Seedance on Yapper
Seedance 2.5 is coming soon. Explore the model page, save these prompts, and start planning your first 30-second generation.
