Seedance 2.5: the practical guide

Seedance films. It does not cut, it does not splice, it does not add effects. Whatever you write, it tries to do with a camera, in one go, without stopping. Every rule below comes out of that one fact.

This guide starts with the work: seven rules, then the formula, then where to start, then seven ready prompts, and at the end what to do when the result is bad.

If you read nothing else, read these seven. Most of your result comes from them.

1. 60 to 100 words: write too little and the model invents the rest itself. Write too much and you start contradicting yourself. Count the words, it really does work.

2. One camera move: one, not "it comes closer, then spins, then pulls away". If you truly need two, write them in one breath: camera tracks low alongside her, then rises slightly.

3. Say separately who is moving: one sentence is about the person, the other is about the camera. She turns slowly toward the window. The camera holds a fixed frame. Mix the two into one sentence and the shot starts shaking. This is the single most common cause of a ruined result.

4. Talk about pace, not numbers: slow, gentle, gradual, smooth, steady. No "22% speed", no "f/2.8" or "85mm". It is the exact opposite of image models, and that is what trips up people coming from Midjourney.

5. Light in every prompt: if you can only add one thing, add light. A woman walking and a woman walking in soft golden hour light are simply different quality shots.

6. Write what you want, not what you do not: inside the prompt, positive description only. Whatever you do not want to see goes at the end, after the word avoid.

7. One fast thing, no more: fast camera plus fast person plus a busy scene is a guaranteed ruined shot. Pick one, keep the rest calm.

Seedance puts the most weight on whatever comes first. So the order is not something you can change.

[Subject], [action], in [environment with lighting],
camera [one movement], style [specific reference],
[duration], [aspect ratio], avoid [constraints]

Who or what is in frame, what they are doing, where they are and what the light is like, how the camera moves, what style. In that order. In English, one paragraph, no lists.

That is the formula for a single moment shot, 5 to 8 seconds. When you need several shots in one generation, the format changes. That one is in Part 4.

1Camera moves

Eight moves the model reads reliably. Pick one per shot.

2Light

Never write "well lit". Say where the light comes from and what it is like.

3Words that ruin the shot

The first ones are editing words. A camera cannot do them, so the model either ignores them or invents its own version. The second ones are empty adjectives that take up space and say nothing.

Instead of speed ramp into slow motion, write her stride slows until it is almost still. Instead of digital zoom punch in, write the camera pushes in close on her hands. Instead of cinematic, write cinematic film tone, warm 35mm grade.

You can have your first shot in ten minutes. Here is the order.

Besides Dreamina there are two other routes. Higgsfield and CapCut are faster and cheaper, good for trying ten variants, but 2.5 is not there yet, only 2.0. The API is what you need when the same prompt has to run across fifty products, or when you are building a content agent for a client.

The first five are single moment shots: short, stable, cheap. The last two are multi shot in one generation, for when you actually need 2.5's 30 seconds. Take one, swap the person and the setting for your brand, and leave the rest of the skeleton alone.

One moment: 5 to 8 seconds

A woman in her late twenties sets a laptop down on a cafe table and opens it in one unhurried motion, then leans back. Morning light comes through a tall window from camera left, warm across the table and soft on her face. Camera holds a fixed medium frame. Cinematic film tone, 35mm, warm natural grade, shallow depth of field, clean negative space in the upper third. 6 seconds, 9:16, avoid jitter, bent limbs, identity drift.

Animate @Image1. Preserve the composition, the label and the exact product colour. A slow light sweep travels across the jar, catching the rim and the edge of the lid. Faint steam lifts from a cup beside it and settles. Camera pushes in slowly and stops. Warm wooden surface, soft diffused window light from camera right, subtle realistic reflections, premium ecommerce style, sharp focus on the label. 6 seconds, 4:5, avoid warped text, extra objects, distorted product shape.

A woman in her early thirties, dark hair tied back, wearing a plain black knit, speaks directly to camera with a calm confident delivery. She gestures once with her right hand on the key phrase. A softly out of focus studio wall behind her. Soft natural window light from camera left with gentle fill on the shadow side. Camera holds a fixed medium close frame. Realistic, natural documentary tone, sharp focus. She says: "The tool is not the advantage. The system is." 8 seconds, 9:16, avoid jitter, bent limbs, identity drift.

Transform @Video1. Preserve the original motion, timing and framing exactly. Change the palette to a cool desaturated winter grade with pale blue shadows and muted skin tones. Add fine film grain across the whole frame. Keep the subject's face and clothing consistent with the source. Keep the existing light direction and the shape of the shadows. Documentary realism, 35mm texture, high detail. 8 seconds, 9:16, avoid identity drift, temporal flicker, warped text.

@Image1 stands behind a marble counter in a small bakery, sliding a tray of pastries forward with an unhurried movement. Shelves of bread recede softly into shadow behind her. Warm practical light from a pendant lamp above the counter, deep amber shadows in the corners. Camera holds a fixed medium frame. Cinematic film tone, 35mm, warm grade, high detail. 6 seconds, 4:5, avoid jitter, bent limbs, identity drift, warped text.

Several shots in one generation

When you make something long on 2.5, do not write one long paragraph. Break it into shots and write the seconds beside each one. One line at the top: total length, number of shots, frame shape. The 60 to 100 word limit does not apply here, but the seven rules do, inside every single shot: one move, light written down, person and camera in separate sentences.

Total: 20s / 3 shots / 16:9

Shot 1 (0s to 6s): A young ceramicist lifts a finished bowl from the wheel, her hands wet with clay. Camera holds a fixed close frame on her hands. Late afternoon light rakes low through a workshop window, dust visible in the beam.

Shot 2 (6s to 14s): She carries the bowl across the workshop at an unhurried pace. Camera tracks slowly alongside her at chest height. Deep warm shadows on the clay walls behind her.

Shot 3 (14s to 20s): She sets the bowl down beside three others on a window bench and steps back. Camera settles into a fixed wide frame. She is backlit by the window, warm rim light on her shoulders.

Cinematic film tone, 35mm, warm grade, high detail.
Avoid jitter, identity drift, temporal flicker.

Total: 15s / 3 shots / 9:16

Shot 1 (0s to 5s): Hands lift a small ceramic cup of coffee from a wooden counter and set it down in front of the camera. Camera holds a fixed close overhead frame. Warm morning window light from the left, steam catching the light.

Shot 2 (5s to 10s): A woman sits by the window of a small cafe, both hands around the cup, looking out at the street. She stays still. Camera pushes in slowly. Soft overcast light through the glass.

Shot 3 (10s to 15s): Close on her face as she takes the first sip and her shoulders drop. Camera holds a fixed frame. Warm side light, soft shadow on the far cheek.

Cinematic film tone, 35mm, warm natural grade, shallow depth of field.
Avoid jitter, bent limbs, identity drift, warped text.

Run this list first. Eight items, half a minute, and a pile of saved credits.

• Word count: is the prompt between 60 and 100 words
• Camera: one move, or three
• Light: did you write where it comes from
• Separation: are the person's movement and the camera's movement in different sentences
• Banned words: no editing words, no empty adjectives
• The negatives: is avoid at the end, not inside the text
• Length: under 30 seconds, and does every shot get more than 4
• Frame shape: written down and matching where you are publishing

Then change exactly one thing

Changing three at once means learning nothing.

1. The shot shakes: leave one move, add a pacing word, and check that avoid: jitter is there.
2. The face changes: add avoid: identity drift and attach an image if you have one.
3. It is dull: change the lighting line only. Nothing else.
4. The frame is a mess: cut the words, remove the least important part.
5. Bad every time: go back to the simplest version, meaning person, action, light, still camera. Then add one thing at a time.

What to do this week

1. Open Dreamina, paste in the first prompt and generate it twice. The goal is one thing: see what a properly written prompt actually looks like as a result.
2. Take one photo of your product or service and run it through the second prompt. That one you can already post on Instagram.
3. In the third prompt, swap the line for one sentence of your own and check the pronunciation. If it works, you now have talking content without a camera.

The numbers and features move fast. When you need to check, the two official places are the Dreamina site and the BytePlus ModelArk docs. Everything else is a retelling.