About video generation

NightCafe can turn a text prompt (and optionally a start image) into a short video. Video generation is an early-stage technology and a little less predictable than image generation — sometimes the results are amazing, and sometimes disappointing — so it may take a few attempts to get a great result.

Current video models

Studio starts new videos on PixVerse V6. You can switch to any current video model in the picker. Provider chips in the picker (including MiniMax and Pruna AI) help you jump straight to a family.

The newest cheap option is P-Video-2 (and P-Video-2 Fast) — 720p or 1080p clips up to 20 seconds on any PRO plan, with optional native audio and start/end frames. P-Video-2 Pro / Pro Fast need Enthusiast+ and Fast Credits. See P-Video-2 and P-Video-2 Pro on NightCafe.

The newest ByteDance model is Seedance 2.5 — 4–30 second clips at 480p, 720p, or 1080p, with an optional start frame and end frame, or up to nine reference images. It needs Enthusiast+ and Fast Credits. See Seedance 2.5 on NightCafe.

The newest MiniMax family is H3 Max, H3 Max Turbo, and H3 Fast — 5–15 second clips with native audio. H3 Max Turbo is available on any PRO plan; H3 Max and H3 Fast need Enthusiast+ and Fast Credits. See MiniMax H3 Max, Turbo and Fast on NightCafe and the NightCafe AI Models Guide.

Writing good video prompts

• Be descriptive: include the details of the scene, the action, and how things should move.
• Specify camera motion: e.g. “dramatic zoom in” or “FPV drone shot”.
• Detail the action: e.g. “teddy bear swimming, arms and feet moving, water splashing”.
• Describe key features: e.g. “white teddy bear, sunglasses, soft fur”.
• Set the scene: e.g. “cloudy sunset near a Caribbean beach”.
• Describe the sound: many current models include native audio. Say what you want to hear (dialogue, ambience, music), or that you want no music.

Using a start image

A start image lets you begin the video from a specific frame instead of random pixels. Tips for choosing one:

• Realistic images tend to work better than cartoon or artistic styles.
• Faces, scenery and interiors have a higher rate of good results.
• If you can easily imagine how an image would look animated, it’s more likely to produce a good video — your intuition is a good guide.
• Some models (including Seedance 2.5) do not accept start images of real people.

End frames

Several current models (including P-Video-2, Seedance 2.5, MiniMax H3 Max / Turbo, and Veo 3.1) let you set a last frame as well as a first frame. End frame options are:

• None — no last-frame constraint.
• Loop — the clip ends where it started, so it can play as a loop.
• Image — upload a custom last frame.

End frames need a start image. After you generate, the creation settings panel shows the Video End Frame you chose. If you hide the start image on a video, other people will not see the end-frame image either.

Camera motion options

Many video models let you choose a camera movement:

• None / Static — no specific motion applied (the AI may still add some movement).
• Pan Left / Right — moves the camera horizontally across the scene.
• Orbit Left / Right — rotates around a focal point for a 3D effect.
• Crane Up / Down — moves the camera vertically.
• Move Left / Right / Up / Down — shifts the camera across the frame.
• Push In / Pull Out — zooms in or out.