Skip to main content
Guide9 min read

Seedance 2.5 Flash Prompt Guide: Draft More Takes for Less

How to prompt Seedance 2.5 Flash: the five-part prompt, testing openings, camera moves and endings as separate takes, first and last frames, reference images and credit costs.

VdoBloom Team

Contents (9 sections)

Seedance 2.5 Flash is the lower-cost tier of Seedance 2.5, so the best way to prompt it is to stop looking for one perfect prompt and run several short takes that each change one thing: the opening, the camera move or the ending. In VdoBloom, Flash makes 4 to 30 second clips with sound at 480p or 720p. A 5-second take at 480p costs 28 credits, about half the 56 credits of the same take on Seedance 2.5. Pick it in VdoBloom's Text to Video or Image to Video.

Quick facts

  • Length: 4 to 30 seconds, with sound.
  • Resolution: 480p or 720p. For 1080p, use Seedance 2.5.
  • Shapes: 16:9, 4:3, 1:1, 3:4, 9:16 and 21:9, or follow the first frame.
  • Inputs: text only, a first frame, a first and last frame, or up to 9 reference images.
  • Not supported on Flash: reference videos and reference audio. Use Seedance 2.5 for those.
  • Cost: 28 credits for 5 seconds at 480p, 61 at 720p. The price is the same with or without sound and shows on the button before you run.

Open Seedance 2.5 Flash in Text to Video

What Flash is for

Flash sits below Seedance 2.5 in price and tops out at 720p. That makes it the right place to work out a shot: which opening holds attention, which camera move suits the subject, how the clip should end. Those are choices you can only judge by watching, and watching costs credits. At about half the price per second, you can afford to look at more versions before you commit.

It is also a good fit for long clips on a budget. Flash goes to 30 seconds in one take, the same as Seedance 2.5.

When you need 1080p, or you want to guide the clip with a reference video or an audio track, move to Seedance 2.5. A prompt that works on Flash is a sensible starting point there, but the two are separate runs and results will differ.

The five-part prompt

Write the prompt in the same order every time: subject, action, camera, light, sound. Each part narrows the next, so the model has less room to invent things you did not ask for.

PartWhat to writeExample
SubjectWho or what, with one detail that identifies itA woman in a green rain jacket
ActionA verb with a start and an endsteps off a tram and opens an umbrella
CameraOne move, plus framingslow push in, medium shot
LightDirection and qualityovercast evening, wet street reflections
SoundWhat the scene sounds likerain on the umbrella, tram bell, no music
Code
A woman in a green rain jacket steps off a tram and opens an umbrella.
Slow push in, medium shot. Overcast evening, wet street reflections.
Rain on the umbrella, a tram bell in the distance, no music.

Flash makes sound with the picture, so the sound line is part of the prompt. If you leave it out, the model chooses for you. "No music" and "no dialogue" are valid instructions.

These are prompts to try, not results from a test session. Results vary from run to run.

Turn one prompt into several takes

Keep four of the five parts fixed and change one. That way, when two takes differ, you know why. Run each take short (4 or 5 seconds at 480p) and compare them side by side.

Test the opening

The first second decides whether a viewer stays. Keep the subject, light and sound, and change only how the clip starts.

TakeOpening line to swap in
A, in motionThe clip opens with her already stepping off the tram.
B, detail firstThe clip opens on a close-up of the umbrella snapping open, then widens to show her.
C, revealThe clip opens on the empty wet street; the tram slides into frame and she steps off.

Test the camera move

Give each take one move. Two or three moves in a short clip read as a glitch.

MoveWords to useGood for
Lockedcamera locked off, no movementAction that should carry the shot by itself
Push inslow push in toward her faceEmotion, a product detail
Orbitslow orbit to the right around herShowing a subject from more than one side
Trackingcamera tracks beside her at walking paceMovement through a place
Crane upcamera rises from street level to above the tramAn ending that opens up the scene

On VdoBloom's Text to Video and Image to Video you can also pick a move from the Direct the shot camera picker and leave the camera line out of your prompt.

Test the ending

Many AI clips simply stop. Write the last beat so the clip lands.

  • Hold: "She stops under the awning and looks up. The camera holds on her face for the last second."
  • Exit: "She walks out of the right side of the frame, leaving the empty street."
  • Loop-friendly: "The clip ends on the empty street, the same framing as the opening."

What a round of takes costs

On VdoBloom, Flash is billed per second, and the rate depends only on resolution.

Length480p720p
4 seconds22 credits49 credits
5 seconds28 credits61 credits
10 seconds55 credits122 credits
15 seconds82 credits183 credits
30 seconds163 credits365 credits

Three 4-second openings at 480p cost 66 credits in total. For comparison, one 5-second take on Seedance 2.5 costs 56 credits at 480p and 126 at 720p. So a sensible order is: test at 480p for 4 or 5 seconds, pick the winner, then run that one at 720p and at full length.

Writing a longer Flash clip

A 20 or 30 second take is not one long sentence. Write it as three beats joined by one camera move.

Code
A baker in a white apron slides a tray of bread out of the oven (beat one).
He carries it to the counter and sets it down, steam rising (beat two).
He tears one loaf open and smiles at someone off camera (beat three).
One slow tracking shot follows him from oven to counter.
Warm morning light from the window. Oven hum, tray on wood, no music.
  • Describe the subject once. After the first clause, say "he" or "she". Repeating the outfit mid-prompt can make the model redraw it.
  • Three beats, not five. Two feels rushed at 30 seconds. Five tends to blur together.
  • One camera instruction for the whole clip.
  • Test the beats short first. Run beat one as a 5-second take before you pay for 30 seconds.

First and last frames

In VdoBloom's Image to Video, Flash can start from your image, or start on one image and end on another. A last frame needs a first frame.

  • With a first frame, do not describe the picture again. The model can see it. Use the prompt for what happens next: the action, the camera and the sound.
  • Match the shape. Set the aspect ratio to follow the first frame so nothing is cropped.
  • With a last frame, describe the path between the two. "The camera pulls back from the cup to show the whole cafe" tells the model how to get from A to B.
  • Keep both frames related. Same place and subject, different moment. Two unrelated pictures give a muddy middle.

Reference images

Flash takes up to 9 reference images. Use them for things words describe badly: a face, a product, an outfit, a location.

  • Say what each image is for. "The woman is the person in image 1. She wears the jacket in image 2. The street is image 3."
  • Use few, clear images. One sharp photo of a product beats five similar ones.
  • Only use photos you have the right to use. For people, that means their consent.

Reference videos and reference audio are not accepted on Flash. If you add them, the tool tells you to use Seedance 2.5 and does not charge you.

Common mistakes

  • Changing everything between takes. If the opening, camera and light all change, you learn nothing from the comparison.
  • Testing at full length. A 30-second 720p take costs 365 credits. Find the look in short 480p takes first.
  • Stacking camera moves. "Orbit, then zoom, then crane up" in 5 seconds rarely works. One move per take.
  • Leaving sound to chance. Name the sounds you want and the ones you do not.
  • Adjectives in place of action. "Cinematic, stunning, epic" gives the model nothing to do. A verb with a start and an end does.
  • Expecting 1080p. Flash stops at 720p. Finish on Seedance 2.5 if you need more, or sharpen the result with the VdoBloom Video Upscaler.

FAQ

What is the difference between Seedance 2.5 Flash and Seedance 2.5?

Flash is the lower-cost tier. On VdoBloom it runs at 480p or 720p for 4 to 30 seconds and takes text, first and last frames and up to 9 reference images. Seedance 2.5 adds 1080p and accepts reference videos and audio, at about twice the price per second at the same resolution.

Does Seedance 2.5 Flash make sound?

Yes. It generates audio with the picture, and the price is the same with or without sound. Describe the sound in the prompt.

How long should a test take be?

4 or 5 seconds at 480p is enough to judge an opening or a camera move. That is 22 or 28 credits.

Can I reuse a Flash prompt on Seedance 2.5?

Yes, the same prompt structure works. The result will not be identical, because it is a different run on a different tier, so check it again.

How long can the prompt be?

Up to 20,000 characters, which is far more than you need. Most good prompts are three to six short sentences.

Run your first round of takes. Open Seedance 2.5 Flash on VdoBloom, write the five-part prompt, and test three openings at 480p.

Related reading: Seedance 2.5 prompt guide: 10 prompt patterns, Seedance 2.5 pricing, Image-to-video prompt examples.

Try it yourself

Video Upscaler

Pick a model, add a photo or a prompt, and see the credit price before you generate. Pay per use with credits that never expire.