Skip to content
Playcut Docs

Wan 2.7: lip-sync, video-edit modes and pricing

Wan 2.7 (wan-2.7) is Alibaba’s video model, served on Playcut through DashScope International. It is the only model on the platform that can lip-sync a still image to an audio file you supply, and one of two models behind Playcut’s video-edit mode (source video + prompt → transformed video, no mask — HappyHorse 1.1 is the other, adding reference-image swaps). It also does text-to-video and reference-to-video with up to 5 reference images. Pick it when you need driving audio, when you want to restyle an existing clip, or when you want a mid-priced generalist. Output costs 40 credits/second at 720p and 60 credits/second at 1080p, on every mode.

Specs at a glance

Section titled “Specs at a glance”
Model idwan-2.7
ProviderAlibaba (DashScope International)
Modestext-to-video, image-to-video (animate, optional driving audio), reference-to-video, video-edit
Aspect ratios (Playcut)16:9, 9:16, 1:1, 4:3, 3:4 — reference-to-video only
Resolutions720p, 1080p
Duration2–15 seconds (all generate modes) · 2–10 seconds (video-edit)
Max reference images5 combined (reference-to-video) · 1 first frame (image-to-video) · 1 source video (video-edit)
Credits40 cr/s @ 720p · 60 cr/s @ 1080p
Audio inputYes — WAV/MP3, 2–30 s, ≤15 MB (image-to-video only)
Text-to-videoOffered since 2026-07-24

How it compares

Section titled “How it compares”
ModelMax durationMax refsMax resolutionCredits/s @ 720p
Wan 2.715 s51080p40
HappyHorse 1.115 s91080p56
Seedance 2.015 s74K21
Imagine Video (classic)15 s7720p20
Veo 3.18 s (per call)34k160

Wan 2.7 is the only row in that table with audio input; it and HappyHorse 1.1 are the two with a video-edit mode.

Modes

Section titled “Modes”

Text-to-video

Section titled “Text-to-video”

Available since 2026-07-24. Tool: generate-video-from-text with model: "wan-2.7". Just a prompt (plus optional negativePrompt); aspect ratio may be any of the five in the specs table, duration 2–15 seconds, 720p or 1080p. Wan’s own prompt guide favors shot-by-shot cinematic prompts (Shot 1 [0–3 s]: …).

Image-to-video (animate)

Section titled “Image-to-video (animate)”

Tool: generate-video-from-image. Needs a prompt and one imageAssetId — the still becomes the first frame.

Optionally pass audioAssetId to drive lip-sync from a real audio track. The audio clip must be WAV or MP3, 2–30 seconds, 15 MB or less. The API truncates the audio to your requested durationSeconds; if the audio is shorter than the video, the remainder of the clip is silent. audioAssetId is rejected on every other model.

This mode has no aspect-ratio parameter. Output aspect follows the input frame — and not exactly, because the encoder forces width and height to multiples of 16. Do not rely on pixel-exact aspect preservation.

Duration: 2–15 seconds, integer.

Reference-to-video

Section titled “Reference-to-video”

Tool: generate-video-from-reference. Needs a prompt and 1–5 referenceImageAssetIds.

References are addressed positionally in the prompt as Image 1, Image 2, and so on — capitalized, a space before the number, no brackets. This differs from HappyHorse 1.1, which uses [Image 1] with brackets. Cross-pasting prompts between the two silently degrades identity adherence.

The 5-reference cap is a combined budget on the provider side: reference images, reference videos and a first frame all count against the same 5. Playcut surfaces images only, so the effective cap is 5 images.

Aspect ratio applies on this mode (16:9, 9:16, 1:1, 4:3, 3:4; default 16:9). Duration: 2–15 seconds, integer. Audio input is not accepted here.

Video-edit

Section titled “Video-edit”

Tool: generate-video-edit. Needs a prompt and one videoAssetId. Wan 2.7 and HappyHorse 1.1 both support this mode on Playcut — HappyHorse additionally takes up to 5 reference images for subject swaps.

Source video requirements: MP4 or MOV, 2–10 seconds, 100 MB or less, with width and height each between 240 and 4096 px. A longer source is a hard reject, not a truncation — nothing in Playcut pre-trims it for you.

Output duration is 2–10 seconds; omitting durationSeconds defaults to 5 rather than inheriting the source length (the provider’s “keep the input length” behaviour is not offered, because the credit hold has to be known before dispatch). Output aspect ratio is derived from the source video; edit requests have no aspectRatio field — sending one is rejected with a 400.

Video-edit bills source + output seconds (since 2026-07-23) — the provider itself charges for both, so Playcut’s hold covers both. If your uploaded source has no stored duration, the hold assumes the 10-second maximum and the difference is automatically refunded once the real source length is measured during generation.

The provider additionally accepts up to 4 optional reference images on video-edit. Playcut does not surface them.

Pricing

Section titled “Pricing”
ResolutionCredits per second5 s clip10 s clip15 s clip
720p40200 cr400 cr600 cr
1080p60300 cr600 cr900 cr

1080p (added 2026-07-24) is available on every mode. Video-edit bills (source + output) seconds at the tier rate, so a 10-second 1080p edit of a 10-second source is (10 + 10) × 60 = 1,200 credits. Animate and reference-to-video bill output seconds. Video-edit bills source + output seconds — a 10-second edit of a 6-second source is (6 + 10) × 40 = 640 credits — because the provider charges for both. When the source’s duration isn’t stored on the asset, the hold assumes the 10-second maximum and auto-refunds the unused seconds on completion.

Limitations & gotchas

Section titled “Limitations & gotchas”
  • Text-to-video is available (since 2026-07-24) — same 5 aspect ratios and 2–15s window as reference mode, via generate-video-from-text with model: "wan-2.7".
  • No interpolation, no extension. Both are rejected before dispatch. Use Veo 3.1 for those.
  • Animate has no aspect-ratio control. Output follows the input frame, snapped to multiples of 16 px.
  • Image N, not [Image N]. Wan reference tokens carry no brackets.
  • Audio is image-to-video only. Reference-to-video rejects the audio role, and every model other than wan-2.7 rejects audioAssetId outright.
  • The driving-audio window is a hard 2–30 seconds. Playcut pre-checks the asset’s stored duration where it has one; assets with no recorded duration slip past the check and fail later at the provider.
  • Video-edit source is capped at 10 seconds and 100 MB. Trim before uploading.
  • Prompt extension is on. Playcut sends prompt_extend: true on all three modes, so the model rewrites and expands your prompt. Output is not strictly literal to the submitted text. Watermarking is disabled on every call.
  • Generation is submit-and-poll. All three modes go through DashScope’s async API, so provider errors surface while polling rather than at submit time.
  • Rate limits. DashScope throughput is shared across Wan, HappyHorse and Qwen voice on Playcut’s connection. Long clips can queue.

How to generate

Section titled “How to generate”

Set model to wan-2.7. The full request flow (session → submit → wait → download) is in the Quickstart.

Terminal window
curl -X POST https://api.playcut.ai/api/v1/tools/generate-video-from-image/execute \
  -H "Authorization: Bearer $PLAYCUT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "workerId": "65f3a1b2c3d4e5f600000001",
    "model": "wan-2.7",
    "imageAssetId": "65f3a1b2c3d4e5f600000003",
    "audioAssetId": "65f3a1b2c3d4e5f600000004",
    "prompt": "She speaks to camera, warm natural expression, slight head movement",
    "resolution": "720p",
    "durationSeconds": 10
  }'

From MCP the tools are the same: generate-video-from-image, generate-video-from-reference, and generate-video-edit. See the MCP install guide.

Section titled “Related”