Create your first video
- Select Video Generation on the home page.
- Choose image-to-video or reference-to-video based on your source material. Seedance models that accept prompt-only requests can start with text.
- Upload the required first/last frames, character images, reference video/audio, or source video, and wait for uploads to finish.
- Set the duration, resolution, and aspect ratio supported by the model.
- Enter a prompt with visible action and camera language, review the live credit estimate, then select Generate.
- In My Videos, preview, download, share, reuse settings, or select Extend Video to continue the result.
For a fast, controlled result, create a strong first frame in Image Generation and animate it with an image-to-video model.
A practical video prompt formula
Use: subject + continuous action + scene change + camera movement + framing + light/mood + style + pace.
Example: On a rainy street at night, a woman in a black trench coat walks slowly toward the camera under an umbrella. Wind moves the coat and neon reflections ripple on the pavement. The camera smoothly pushes from a wide shot to a medium close-up. Shallow depth of field, blue-violet cinematic light, natural motion, calm pace.
Keep these rules in mind:
- Use observable verbs such as turns, looks up, or walks toward the camera.
- Give a short shot one primary action; too many actions often cause jumps.
- Name the camera movement: locked, push in, pull out, pan, orbit, tracking, or handheld.
- For image-to-video, focus on what happens next instead of repeating the visible appearance.
- When using multiple references, bind each subject with the reference notation shown in the panel.
Choose a workflow
| Workflow | Best for | Prompt focus |
|---|---|---|
| Text to video | Exploring a shot from scratch | Subject, setting, action, camera, and style |
| Image to video | A character/product first frame is ready | Motion, camera movement, and environmental change |
| Reference to video | Multiple characters, objects, video, or audio references | Each reference's identity and role; bind every image when the panel asks for reference notation |
| Video extension | Continuing from an existing ending | Use Extend Video in history to extract the final frame and continue from it |
Limits are model-specific: current reference models accept up to 3–10 images and up to 3–9 videos or audio files. Videos must be MP4/MOV up to 100 MB; audio must be MP3/WAV up to 15 MB. The uploader shown for the selected model is the source of truth.
Video model guide
| Model | Main capability | When to choose it |
|---|---|---|
| Wan 2.7 Spicy / Uncensored | Image-to-video or reference-to-video | Choose the variant for your source material |
| Wan 3.0 / WAN 3.0 Prime | Image and multi-reference video | Longer duration options, several resolution tiers, image/video/audio references, and model-specific audio controls |
| Seedance 2.0 / 2.5 Spicy | Image and reference video | Model-specific last-frame, generated-audio, or adaptive-ratio support |
| MiniMax H3 / Enhanced / Max | Image or reference video | Flexible durations, resolution tiers up to 4K, and multi-image/video/audio reference support depending on the variant |
Duration, resolution, and credit rates vary by model. Use the valid options displayed after selecting a model.
Improve stability
- Start from a sharp frame with a clear, unobstructed subject.
- To control the ending, choose a model that exposes Last Frame and upload the target image.
- Begin character shots with subtle movement, then increase motion gradually.
- If the camera shakes or anatomy drifts, request a stable camera, continuous motion, and stable subject structure.
- Split a longer story into single-action shots, then combine them with AI Director or an editor.
- If Generate is unavailable, check prompts, required uploads, reference bindings, model/mode compatibility, and available credits.
Extend a finished video
In My Videos, select Extend Video on a completed result. 可梦KDream extracts its final frame and sets that frame as the next video's first frame. Choose an available image-to-video model and describe the next action and camera move.
