Dreamina Image to Video Guide: Seedream, Seedance, Settings

Dreamina (dreamina.capcut.com) is ByteDance’s generation suite, sitting alongside CapCut. It makes images with its Seedream models and short videos with its Seedance models, from text or from an image you upload or have just generated. For a working creative the practical path is image first, video second: get a still exactly right, then animate it. This guide covers that two-step workflow, the settings that matter at each stage, and how to hand the result to CapCut for editing.

How Dreamina is organised

The interface separates image generation and video generation, with a model picker in each. On the image side you will see Seedream versions; on the video side, Seedance versions. Newer versions generally follow prompts more closely and handle text and hands better, but they can cost more credits. Both tools share the same account credits, which refill according to your plan. Check the platform’s current pricing page for the details, because the allowances change.

Dreamina also connects to CapCut, so a generated clip can go straight into a CapCut project for trimming, captions and music. That matters if your output is short-form social video, where the generated clip is rarely the finished product.

Step one: a still worth animating

Video inherits every flaw in the start frame, so spend time here.

1. Choose the aspect ratio for the final video. 9:16 for vertical social, 16:9 for landscape. Generating square and cropping later wastes resolution.
2. Pick the Seedream model and quality. Use a cheaper or faster option to explore ideas, then regenerate the chosen composition at the highest quality available.
3. Write a scene prompt. Subject, setting, light, composition, in two or three sentences. Leave motion out; it belongs in the video step.
4. Leave room for movement. If the subject will walk or the camera will push in, do not frame them tight against the edge.
5. Check faces, hands and text at full size. Fix them now with another generation or an edit rather than hoping the video model will.

For prompt structure in detail, see the Dreamina prompts guide.

Step two: image to video with Seedance

Send the still to the video tool (or upload one). Then:

• Write a motion prompt, not a description. The model already sees the image. "The woman turns towards the window and the curtain lifts in a breeze. Slow push-in." is enough.
• One action per clip. Short clips cannot hold three events. Pick the one that matters.
• Name the camera separately. "Static camera", "slow pan left", "gentle handheld". If the interface offers camera presets, keep the text consistent with your choice.
• Calm the motion with adverbs. "Slowly", "gently" and "subtle" reduce warping, which is the most common failure in short clips.
• Choose duration and quality. Start with the shortest duration and standard quality to test the idea, then rerun at higher quality for the keeper.

Review the first and last second frame by frame. That is where artefacts tend to appear, especially around hands, hair and anything with lettering.

Settings at a glance

Labels in the current interface may differ slightly from these names, and model lists change often, so check what is selected before each run.

Common problems

• The subject melts or stretches. Too much motion for the duration. Simplify to one action and add "subtle".
• Nothing moves. The prompt described the scene again. Use a verb and a camera move.
• Text in the image smears. Video models struggle with lettering. Remove it from the still and add titles in CapCut.
• Colour shifts between still and clip. Mention "colours stay consistent" in the motion prompt and avoid lighting changes like "sun sets".
• Credits run out mid-project. Test at standard quality and short duration, and only upgrade the shots you keep.

Try it

Image prompt (Seedream): A glass teapot on a wooden counter, steam rising, soft window light from the left, shallow depth of field, 9:16.
Video prompt (Seedance): Steam drifts slowly upward and thins. Light shifts gently as a cloud passes outside. Slow push-in, static background, colours stay consistent.
Duration: shortest. Quality: standard for the test.

Once the test works, rerun the video at the highest quality and swap "slow push-in" for "static camera" to see how much of the motion comes from the scene versus the lens.