Wan 2.7 AI Video Generator

Three capabilities set it apart from its siblings. First, keyframe control: supply a first frame, a last frame or both, and the model works out the motion path between them while keeping the subject stable. Wan 2.6 only accepts a first frame.

Second, multimodal references. Beyond images, Wan 2.7 can take an audio clip and a reference video. On Cinel you can add up to 3 reference images to keep characters and visual details consistent, and the model can match a character's speech to the voice in a 1–10 second sample — useful for recurring hosts or spokescharacters.

Third, finer prompt control: negative prompts tell the model what to avoid (extra fingers, text overlays, a cluttered background), and optional prompt enhancement expands short ideas into fuller briefs. At model level Wan 2.7 also supports instruction-based editing of an existing 2–10 second clip, such as swapping the background for a rainy street without regenerating everything.

• First and last frame control — plan exactly where the shot begins and ends.
• Image, audio and video references — up to 3 images plus a 15 s audio or video reference on Cinel (audio/video for members).
• Voice matching — character speech follows the vocal style of a short audio sample (model capability).
• Audio-driven timing — attach music or a voiceover and the motion is paced to it; if the track is shorter than the clip, the remainder is silent.
• Negative prompts and prompt enhancement for cleaner results.
• Any length from 2 to 15 seconds.
• 720p or 1080p (1080p ≈ 1.67× per second).
• Model-level video editing of existing short clips.

Item	Wan 2.7
Developer	Alibaba — Tongyi Wanxiang team
Modes	Text-to-video, image-to-video with first/last frame, reference video; model-level video edit
Duration	2–15 s (Cinel default 5 s)
Resolution	720p / 1080p at 30 fps (1080p ≈ 1.67× per second)
Aspect ratios	16:9, 9:16, 1:1, 4:3, 3:4
References on Cinel	Up to 3 images · 15 s audio · 15 s video (audio/video member-only)
Voice sample	1–10 s for voice matching
Prompt tools	Prompt up to 5,000 characters; negative prompt up to 500; optional prompt enhancement
Cinel cost	From 160 credits

1. Select Wan 2.7 in the video generator (video requires a membership or credit pack).
2. Choose your control method: upload a first frame, a first and last frame, or a set of reference images.
3. Members can add a short audio clip to set a character's voice or a video clip for motion and style guidance.
4. Write the prompt, then add a negative prompt listing what to avoid — "no subtitles, no extra people, no warped hands".
5. Set duration (2–15 s), aspect ratio and resolution; review at 720p, then render the keeper at 1080p.

• Pair first and last frames that share lighting and style, and describe the movement that connects them.
• Write negative prompts as short comma-separated items rather than sentences.
• Keep voice samples clean: a single speaker, no music, within the 1–10 second range.

• Recurring spokescharacters — the same host look and voice across a series.
• Precise transitions for ads — open on a product, end on a pack shot you supply.
• Explainers with a consistent narrator — voice-matched dialogue from a sample.
• Music-synced social clips — upload a short track and let the motion follow its rhythm.
• Fashion and beauty looks — multiple reference images for outfit, makeup and setting.
• Background and setting changes — model-level edits of short existing clips.

Compared with Wan 2.6 (from 145 credits), Wan 2.7 adds last-frame control, audio and video references, voice matching and negative prompts; choose Wan 2.6 when automatic multi-shot storytelling matters more. Compared with Kling O1 (from 230 credits, 5/10 s only), Wan 2.7 offers first-and-last-frame animation at a lower starting price with any duration from 2 to 15 seconds, plus the option to drive the clip with your own audio track.