Seedance 2.5 is the best video model available right now. We still believe that, and most of the agencies building on Chromatic would say the same thing. It is also the reason we spent six months and around $4,000 re-running shots that a cheaper model would have delivered correctly on the first pass.
When Seedance launched, it became our default for everything. UGC videos, cinematic shots, speaking videos, product films, motion graphics. On our own agency work and in the defaults we handed to teams running their production through the platform. In tests it looked excellent, so nothing prompted us to question it.
Production is where it broke, and it broke on the simplest shots in the brief.
It does not reliably know left from right
We were making a film for a home furnishings brand. Three people in a queue at a checkout counter, each pushing a trolley loaded with décor. The motion is simple : one walks in, waits in the centre, pushes out, and the next one takes the place. Static camera throughout.
We wrote a detailed prompt with motion direction, beats and movement described for every person in the frame.
[Image1] is the first frame and primary scene composition lock — woman in olive green velvet puff-sleeve mini dress, gold strappy heels, holding a brown beaded shoulder bag, pushing a chrome shopping trolley filled with luxury décor (beaded cushion, gold metal leaf sculpture, dark horse statue, ceramic vases, gold candle holder). She stands against a deep red wall with dark wood wainscoting, glossy brown floor reflecting softly. [Image2] is the reference image for the second character — clean-shaven man in dark navy blazer, white tee, navy trousers, white sneakers, gold bangles on right wrist, pushing an identical chrome trolley filled with a beige tasselled cushion, pleated lampshade, amber ribbed glass jar with gold pineapple lid, ceramic soap dispenser, tan textured vase, woven planter, ornate gold scrollwork piece. [Image3] is the reference image for the third character — woman in floor-length silver sequinned long-sleeve gown, purple velvet platform heels, stacked gold bangles, holding a smartphone in her right hand, pushing an identical chrome trolley filled with a tan circle-pattern cushion, olive green suede lampshade, olive glass bottle vase, gold horse head sculpture, white ceramic swan dish, small glass of milk with coffee beans. All Image1, Image2, Image3 are standing in queue and will appear one by one in order Image1 olive green dress girl, Image2 drak navy blazer boy, Image3 long sleeve gown girl. Single locked-off static camera, side-on three-quarter angle on the queue, framed against the deep red wall, dark wood wainscoting, glossy brown reflective floor. The camera does not move at any point — fixed framing, constant focal length, constant distance from the scene, single continuous uncut shot. The subjects move; the camera stays still. Mood: three party-bound shoppers waiting in a luxury homeware checkout queue, prepping for a celebration tonight, a thumping party track audible in the background — confident, playful, excited energy. Each person's body grooves subtly to the beat while they wait their turn. Beat 1 (0s–4s): Woman from [Image1] stands centre-frame with her loaded trolley, both hands on the handle, exactly as Image1. Her hips sway gently to the party beat, weight bouncing rhythmically between her gold heels — a clear visible groove, heels tapping the glossy floor in time, knees giving small bounces, shoulders rolling subtly with the rhythm. Beaded shoulder bag sways with each beat. Her smile reads anticipation. Trolley contents stay completely still and lock exactly to reference — beaded cushion, gold leaf sculpture, dark horse statue, ceramic vases, gold candle holder all preserve their exact shapes, materials and positions throughout. Beat 2 (4s–5s): She pushes her trolley forward, move out of frame with her cart. As she clears the frame, the man from [Image2] rolls his loaded trolley in and settles into her vacated centre-frame position, both hands on the handle. Beat 3 (5s–9s): Man from [Image2] stands centre-frame with his trolley, . He nods his head to the party beat, weight rocking heel-to-toe in his white sneakers, one foot doing a clear tip-tap on the floor in sync with the rhythm. Knees bounce subtly, blazer shoulders move with each beat, gold bangles glint as his right hand taps the trolley handle in time. Confident, into-the-music demeanour. Trolley contents preserve exactly to [Image2] reference — tasselled cushion, pleated lampshade, amber ribbed jar with gold pineapple lid, soap dispenser, tan textured vase, woven planter, gold scrollwork piece all hold their exact forms, materials and positions. Beat 4 (9s–10s): Man pushes his trolley forward and move out of frame with his trolley. Woman from [Image3] glides her loaded trolley in from screen-right into the vacated centre position. Beat 5 (10s–15s): Woman in silver sequinned gown stands centre-frame with her trolley, left hand on the handle, right hand holding her smartphone at chest height. She glances between her phone and the queue ahead, hips swaying smoothly to the party beat, purple velvet platform heels stepping a soft side-to-side groove on the glossy floor, sequins catching the light and shimmering with each rhythmic movement, stacked gold bangles glinting on her wrist as her shoulders roll subtly with the music. Playful, party-ready energy. Trolley contents preserve exactly to [Image3] reference — circle-pattern cushion, olive suede lampshade, olive glass bottle, gold horse head sculpture, white swan dish, milk glass with coffee beans all hold their exact forms and positions throughout. Throughout the full 15 seconds: the deep red wall, dark wood wainscoting and glossy brown reflective floor stay completely consistent. Lighting holds steady — soft warm editorial studio lighting with gentle highlights along the floor reflection. Photorealistic editorial commercial style, smooth natural human motion, visible rhythmic groove on each character's lower body and shoulders, stable framing, smooth cinematic motion, consistent lighting throughout, single continuous uncut shot. All outfits, accessories, jewellery and trolley contents stay sharp, proportionally correct, and identical to reference images at all times — never warping, melting, or shifting shape.
Seedance had everyone pushing their trolleys backward, and a few of them entered the frame from the opposite side entirely. The movement looked unnatural in a way that no amount of prompt rewriting fixed.
We ran the same prompt and the same images through Kling V3 Pro.
It came back exactly as directed, with full prompt adherence on the first attempt.
This is three people walking in a straight line. It is the least demanding thing in the entire film, and it is where the most capable model on the market kept failing. We started seeing the same pattern reported by agencies on the platform, on queues, walk-throughs, product handoffs, anything where the direction of travel actually matters to the story.
Complex dialogue outside English falls apart
We animated an illustrated character and gave her one line of formal Hindi to speak.
"मुझे रक्षा नहीं, अधिकार चाहिए। जिस राज्य की राजकुमारी अपने ही आँगन में असुरक्षित हो, वहाँ सिंहासन नहीं, मर्यादा के नाम पर बंधन होता है।"
Words like अधिकार and मर्यादा sit outside daily spoken Hindi. They belong to a more formal register, and the consonant clusters in them are harder to articulate.
Seedance handled the performance well. The expression matched the line, the camera movement was right, the timing felt natural. The pronunciation was where it came apart, and it came apart precisely on the difficult words.
Use the provided image as the exact first frame. A vertical (9:16) close-up-to-medium portrait video in the same painterly, semi-stylized digital illustration style as the source — soft brushstroke background, clean expressive linework on the character, warm golden-hour dappled forest light. Preserve her exact identity, face, braid, jewelry, sari colors and patterns, and the forest backdrop throughout every frame. She is a brave rajkumari (princess) standing in the same dense golden-lit forest clearing, hands clasped in front of her at first. Camera holds a steady, slow, almost imperceptible push-in toward her face over the full duration — no cuts, one continuous shot. Environment motion: the surrounding leaves and thin tree branches sway gently in a light breeze throughout, loose strands of her hair and the sheer blue pallu drifting slightly with the same air movement. The dappled sunlight filtering through the canopy shifts and flickers subtly on her face and shoulder as the leaves move, so the warm highlight on her cheek and jaw brightens and softens in small waves rather than staying static. Emotional arc — she is fierce, brave and dignified, never losing her beauty or grace even in anger: 0:00–0:04 She begins calm but resolute, chin lifted slightly, eyes steady and clear, a faint hardening at the brow. Her lips begin to move, voice measured and firm. Dialogue: "मुझे रक्षा नहीं, अधिकार चाहिए।" 0:04–0:05.5 A short deliberate pause. Her jaw sets, her gaze sharpens further, eyes narrowing slightly with quiet intensity, breath visible in her shoulders, the light on her face dimming slightly as if a cloud-shadow of emotion passes. No dialogue in this pause. 0:05.5–0:10 Her eyes shift further into bravery and controlled anger — brows drawing in, eyes glinting hard and unflinching, a subtle tightening around her mouth, her clasped hands tensing slightly at her waist. Her voice rises in conviction, still dignified, never shouting. Dialogue: "जिस राज्य की राजकुमारी अपने ही आँगन में असुरक्षित हो," 0:10–0:14 She delivers the final line with full fierce resolve, eyes blazing with controlled fury and pride, brow low and intense, lips firm, chin lifted in defiance, her stance subtly straightening as if standing taller. Dialogue: "वहाँ सिंहासन नहीं, मर्यादा के नाम पर बंधन होता है।" Her expression throughout carries bravery, fierce dignity and simmering anger in her eyes, while her face remains beautiful, graceful and regal — never distorted or exaggerated, always composed strength rather than rage. Natural Hindi lip-sync matching the exact dialogue and the timing of both pauses. Small natural human details preserved even in the illustrated style: slow deliberate blinking, visible breathing in the chest and shoulders, subtle micro-tension in the jaw and brow, unhurried and commanding pace. Hair and pallu fabric respond naturally to the breeze, floating and settling realistically. Background foliage, tree trunks and light patches remain the same location throughout, only animated by gentle wind and shifting dapples of golden light — no scene changes, no new elements introduced. No music, no captions, no on-screen text. Audio: only her voice, strong and emotionally charged, with faint ambient forest sound — leaves rustling, distant birds — during the pauses.
Grok Imagine 1.5 got the words right. Pronunciation was accurate, the emotion held through the line, the camera movement worked and the lighting stayed clean.
Seedance is dependable on conversational local language. The moment a script moves into formal diction or heavier vocabulary, the words stop landing, and for brand work in Indian languages that is most of the script.
You are overpaying for multi-shot sequences
This one came out of a jewellery campaign. One model wore a stud earring, two layered necklaces, a bracelet and two rings on the same hand.
We needed a short film built from that single frame. A closeup on each piece, a wide, a medium, different expressions across the cut, all inside one video, with a little drama in the light.
Six months earlier we would have sent this to Seedance without a second thought. We ran it on Minimax instead. It cost half as much and came back about twenty times faster.
R2V. @Image1 is the identity and wardrobe anchor — a woman with wavy dark hair, warm skin, oval face, wearing a sage-green silk cami top, gazing softly upward and to the side; preserve her face, hair and top exactly, only adding motion and expression. Her skin renders bright, smooth and clean with soft pink blush makeup on the cheeks and glossy natural lip tone — no visible pores, no texture, no blemishes. Lighting stays mostly bright and even; any shadow present must be soft and minimal, sculpting the cheekbone or jaw gently, never darkening or roughening the skin — the shadowed side stays as smooth and clean as the lit side. @Image2 is a long silver necklace, vertical pendant: round peridot above blue-topaz marquise drop. @Image3 is a shorter silver necklace: blue-topaz marquise beside peridot bezel, layered higher on the throat. @Image4 is a pair of silver drop earrings, peridot marquise above blue-topaz round. @Image5 is an open silver bangle with peridot-topaz-peridot bezels. @Image6 is a silver interlaced knot ring. @Image7 is a silver pavé eternity ring. Match every gemstone shape, color, chain gauge, setting and metal tone from these references exactly — do not distort or invent jewelry. Editorial luxury look: bright soft key light from camera-left, clean minimal shadow, unhurried elegant motion, fast confident cuts between beats. 00:00-00:01.5 — Her hand rises past her jaw, @Image6 and @Image7 catching the bright key light as they cross her fingers. 00:01.5-00:03 — Cut to her wrist turning, @Image5's topaz center stone glinting against smooth bright skin. 00:03-00:04.5 — Cut to her earlobe, @Image4 swaying gently as she tilts her head, soft sparkle catch. 00:04.5-00:06.5 — Cut to a close-up on her throat, @Image2 and @Image3 layered together settling on her collarbone as she breathes in, every stone lighting up in unison. 00:06.5-00:08 — Cut to her gaze lifting slowly toward camera. 00:08-00:09.5 — Cut to a soft, unhurried smile forming at the corners of her mouth. 00:09.5-00:11 — Cut as she tilts her head the other way, hair shifting to reveal the second earring of @Image4. 00:11-00:13 — Cut to her hand lowering to rest near her collarbone — @Image5, @Image6, @Image7, and both @Image2 and @Image3 necklaces all visible together in one composed frame. 00:13-00:15 — Final held close-up, chin lifted, the full look from @Image2 through @Image7 glowing together in bright even light, gentle fade to black. Music: a full, present light editorial jewelry score throughout — soft plucked strings and delicate piano with a gentle airy shimmer/bell layer, slow luxurious tempo, subtle swell as the necklaces settle at 00:04.5, soft resolving cadence on the final fade. No ambient noise-only track — the music should feel intentional and audible the entire 15 seconds, not muted or buried. No dialogue. Constraints: no text, no subtitles, no watermark, no logos or brand marks anywhere in frame; do not add jewelry beyond @Image2 through @Image7; skin stays bright, smooth, clean and high quality throughout — zero visible skin texture, pores, or flaking, even in shadowed areas.
We make a lot of these. If you want to see what the finished films look like, they are here.
Every piece held its shape across all eight shots, with the model's expression, the camera movement and the light drama intact. Seedance would very likely have produced this too. It would have taken longer and cost twice as much for an output the client could not have told apart.
What we do now
We stopped having a default.
Every model carries its own strengths and its own blind spots, and the useful question is never which model is best. It is which model is best for this shot, at this budget, on this timeline. Answering that question correctly forty times in a campaign is where the margin actually lives.
So we organised what we learned into a knowledge base and gave it to Light, the assistant that runs on Chromatic. Light reads the brief, understands the context of what is already on the canvas, and picks the model based on what the shot needs, whether that is spatial accuracy, dialogue in a specific language, cost or latency.
Hundreds of videos run through Chromatic every day, across our own agency work and the agencies producing on the platform. Every one of them tells us something about where a model holds and where it slips, and that goes back into how Light routes the next brief. The knowledge compounds, and the work gets cheaper and faster as it does.
