This guide assumes you already know the basics of generating a clip in Seedance 2. Here we focus on the upload workflow, motion prompts, and the failure modes that ruin product footage.

Why use image-to-video instead of text-to-video?

Because consistency. Text-to-video invents a new product every run, which is useless when you need the same bottle, sneaker, or watch across ten clips. Image-to-video locks the first frame to your asset, so the object stays recognizable from the first millisecond.

A realistic use case: you shoot one clean product photo on a white sweep, then generate 5-second clips with different camera moves for a catalog page. Ten clips, one shoot day.

What image should you upload?

Resolution matters less than clarity of the subject. Upload a photo where the product occupies at least 40% of the frame and the edges are sharp. Blurry inputs produce blurry motion.

• Use PNG or high-quality JPG. Seedance 2 accepts both, but compression artifacts get animated along with everything else.
• Avoid busy backgrounds. A patterned backdrop will writhe and wobble during the clip.
• Match aspect ratio to output. Uploading a square photo and selecting 16:9 gives the model cropping decisions you will not like.

How do you prompt motion on a still image?

When the first frame is fixed, your prompt becomes a camera and physics instruction. Describe how the scene moves, not what is in it.

Good motion prompts read like this: “slow 30-degree orbit around the object, soft key light from the left, subtle reflection on the surface.” Bad ones re-describe the product, which the model can already see.

1. Pick one movement: orbit, push-in, tilt, or pan.
2. State the light behavior if it changes.
3. Add “no morphing, no text” to the negative field.
4. Keep duration at 5 seconds — beyond that the model starts inventing new geometry.

What are the common failure modes?

The most frequent one is edge warping, where flat product surfaces start to breathe and ripple. It shows up most on glossy packaging. Lowering the movement intensity helps.

The second is logo drift: small printed text on a label smears after the two-second mark. If the logo matters, keep the clip to 5 seconds and use a gentle push-in rather than an orbit.

A third issue is cost. Image-to-video with a 1080p 12-second output burns credits roughly two and a half times faster than a 5-second test. Iterate at 5 seconds, then do one final pass at full length.

Who is this mode not for?

If your product is transparent or highly reflective — glassware, jewelry, chrome — expect cleanup work in a video editor afterward. Seedance 2 handles matte surfaces far better. For character animation or anything needing speech, this is the wrong tool entirely.

Quick Summary

• Image-to-video keeps your exact product consistent across many clips.
• Upload a sharp photo with the product filling 40%+ of the frame.
• Prompt camera motion and lighting, not the object itself.
• Keep clips at 5 seconds to avoid geometry drift and credit burn.
• Glossy and reflective products need post-editing cleanup.

The catch is that most people write prompts like search queries, then blame the model when the result looks flat. This walkthrough covers the signup, the prompt shape that works, and the export settings you will actually want.

What is Seedance 2?

Seedance 2 is a text-to-video and image-to-video model from ByteDance’s Seed team. It outputs clips between 5 and 12 seconds at up to 1080p, and it can animate a still image you upload as the first frame. There is a free tier with a small monthly credit pool, and paid plans start around $10/month for heavier use.

It is not a full editing suite. No timeline, no audio layering, no multi-shot stitching. You generate a clip, download it, and assemble everything elsewhere.

How do you get started?

Sign in with an email or ByteDance account, then land on the generation screen. The interface has three inputs: a prompt box, an aspect ratio selector (16:9, 9:16, 1:1), and a duration slider.

1. Claim the free credits. New accounts get a starter balance, enough for roughly 6 to 10 short clips.
2. Pick your aspect ratio before writing. A vertical 9:16 prompt reads differently than a wide 16:9 one.
3. Set duration to 5 seconds for your first test. Cheaper, faster, and easier to judge.
4. Generate and wait. Queue times sit between 60 and 120 seconds on the standard tier.

How should you write the prompt?

Seedance 2 responds best to a described shot, not a list of keywords. Give it a subject, an action, a camera move, and a lighting condition. Compare “cat” with “a ginger cat walking across a sunlit windowsill, slow dolly-in, warm afternoon light.” The second one gives the model something to animate.

Three things that reliably improve output:

• Mention one camera move only. Two competing moves usually produce mud.
• Name the light. “Neon reflections on wet asphalt” beats “night.”
• Keep it under 60 words. Longer prompts tend to dilute the subject.

Negative prompts are supported too. Add “no text, no watermark” if you keep seeing garbled lettering in the corners.

Is the free plan enough?

For testing, yes. For anything you plan to publish weekly, no. The free credits refresh slowly, and rerenders eat them fast. A 12-second 1080p clip costs noticeably more than a 5-second one, so budget your longer attempts.

Who should skip Seedance 2 entirely? Anyone needing clips longer than 12 seconds, or anyone who wants dialogue synced to mouth movement. For that, look at a dedicated talking-avatar tool instead.

How do you export and use the clip?

Every finished generation sits in your library as an MP4. Hit download and you get the file at the resolution you picked. There is no built-in upscaler, so if you need 4K you will need external software.

One practical note: generate two or three variations before you commit. Seedance 2 is not deterministic, and the same prompt can land very differently on the second try.

Quick Summary

• Seedance 2 makes 5 to 12 second 1080p clips from text or a first-frame image.
• Free tier gives a small credit pool; paid plans start around $10/month.
• Write a described shot with one camera move, not a keyword list.
• No audio, no timeline, no upscaling — it is a generator, not an editor.
• Best for short social clips; skip it if you need clips over 12 seconds.