Playcut v2 is coming MCP · Actor v2 · & much more
alibaba wan 2.7 · 40 cr per second at 720p · video edit

Wan 2.7: make a still speak your line,
or re-cut a clip by prompt

Record the line, pick the still, and Wan 2.7 animates the image to your audio track. Or hand it a short clip and a sentence describing the change, and it re-cuts the footage. Wan 2.7 pricing on Playcut is 40 credits a second at 720p and 60 at 1080p on every mode, with one twist: video edit bills the source seconds as well as the output.

720p and 1080p · 2–15 s · audio-driven image to video · video edit · up to 5 references · rates in credits

Verdict

Is Wan 2.7 the right model for the job?

Wan 2.7 sits in the middle of our video price list, and two jobs are the reason to pick it: performing a still to audio you already have, and editing a short clip by prompt. Veo 3.1, our default video model, does neither.

Reach for it when

  • The voice track exists already, a recorded read or a voiceover, and a still has to perform it.
  • You want to restyle or re-cut a 2 to 10 second clip by describing the change.
  • A single 15-second shot at 720p for 600 credits fits the budget.
  • Your references fit in five images.

Look elsewhere when

  • You need 4K. Wan stops at 1080p.
  • You need to extend a clip or bridge a first and a last frame. Wan has neither mode on Playcut.
  • Your edit source runs past 10 seconds. It is rejected, not trimmed.
  • You need more than five references. HappyHorse 1.1 and Seedance 2.0 take nine.
The money part

Wan 2.7 pricing, mode by mode

Every Wan mode bills at the same two rates: 40 credits a second at 720p and 60 at 1080p. What changes by mode is which seconds count. A failed or cancelled job gets its held credits back in full.

Mode Seconds billed 5 s out, 720p 10 s out, 720p 10 s out, 1080p
Text to video Output seconds 200 cr 400 cr 600 cr
Image to video, with or without driving audio Output seconds 200 cr 400 cr 600 cr
Reference to video Output seconds 200 cr 400 cr 600 cr
Video edit (source the same length as the output) Source + output seconds 400 cr 800 cr 1,200 cr

How video edit billing works

Video edit is the one mode where the input costs money. The provider charges for the clip it reads and the clip it writes, so Playcut's hold covers both: source seconds plus output seconds, times the rate.

A 6-second source edited into a 5-second output at 720p is (6 + 5) × 40 = 440 credits. A 10-second source into a 10-second 1080p output is (10 + 10) × 60 = 1,200. If the source has no stored duration, the hold assumes the 10-second maximum and the difference is refunded once the real length is measured.

Output length does not follow the source. Leave it unset and you get 5 seconds; set it anywhere from 2 to 10. A shorter output is the easiest way to trim the bill on a long source.

What a month of credits buys on Wan 2.7

Plan Credits / month 720p seconds 1080p seconds 10 s edits at 720p Jobs at once
Hobby $9/mo 500 12s 8s 0 1
Pro $29/mo 2,000 50s 33s 2 3
Studio $79/mo 6,000 150s 100s 7 6
Agency $149/seat/mo 10,000 250s 166s 12 10

The edit column assumes a 10-second source edited into a 10-second output, 800 credits each. Hobby's 500 credits cover 12 seconds at 720p, enough for one talking shot rather than a campaign.

If a month runs short, a credit pack (600, 2,500 or 5,000 credits) tops it up and never expires. What each plan includes is on the Playcut pricing page. To cost a batch of talking shots before you commit, use the AI video credit calculator.

In the studio

How to use Wan 2.7 in Playcut

Open the video studio at app.playcut.ai and switch the model picker from Veo 3.1 to Wan 2.7. Nothing reroutes the job to another model behind your back. Then follow the recipe for the job you have.

Make a still speak your recorded line

  1. 1. Prepare the audio. Export the line as WAV or MP3, between 2 and 30 seconds and no larger than 15 MB.
  2. 2. Choose image to video. Add the still as the first frame and attach the audio. There is no ratio setting on this mode: the output follows the image's shape, snapped to multiples of 16 pixels.
  3. 3. Match the length to the line. The clip runs 2 to 15 seconds. Audio longer than the clip is cut at the clip's end, and audio shorter than it leaves the rest silent.
  4. 4. Direct the performance in the prompt. Describe expression, head movement and framing. The audio supplies the words.

Time the read before you record it. The video script timer estimates a script's runtime at slow, normal and fast speaking pace, which tells you the clip length to ask for.

Re-cut an existing clip by prompt

  1. 1. Trim the source first. It has to be 2 to 10 seconds, MP4 or MOV, 100 MB or less, and between 240 and 4,096 pixels on each side. Nothing trims it for you.
  2. 2. Choose video edit and describe the change. For example: "turn the street into a snowy evening, keep the camera move and the cyclist". The output keeps the source's shape.
  3. 3. Pick an output length from 2 to 10 seconds. Remember the source seconds are billed too.

Naming references in a Wan prompt

Reference to video takes up to five images, named in the prompt as Image 1, Image 2 and so on: a capital I, a space, and no brackets. HappyHorse 1.1 writes the same thing as [Image 1] and Seedance 2.0 as @Image1. Pasting a prompt across models without changing the tokens weakens how well identity holds.

The red vintage scooter from Image 1 rolls down the cobbled hill street from Image 2, the camera tracking alongside at wheel height, late afternoon sun.

Your prompt gets expanded before it renders

Playcut sends every Wan request with Alibaba's prompt expansion switched on, so the model rewrites and expands your prompt before rendering. Put the details you care about plainly and early, and do not expect a literal reading of every word. The guide to fixing a weak AI video prompt lists the other model rules that change a prompt before it renders.

That is why the Wan 2.7 prompts for talking photos, scenes and edits are written short, with the details that matter stated first.

From code, over MCP or REST

Pass model: "wan-2.7" to the text, image and reference video tools. Wan 2.7 is already the default model on the video-edit tool. Generate a key and connect a client from the MCP server and REST API page.

The honest part

Where Wan 2.7 stops: no 4K, 10-second edit sources

The limits below are the ones that decide whether Wan fits a brief. When a job fails on one of them after credits are held, the hold is refunded.

No 4K
Wan renders 720p and 1080p only. Seedance 2.0 and Veo 3.1 are the 4K routes we run.
No interpolation and no extension on Playcut
Alibaba's Wan 2.7 image-to-video documentation describes first-and-last-frame and continuation tasks. Playcut does not run either on Wan, and both are rejected before dispatch.
The driving-audio window is hard
2 to 30 seconds, WAV or MP3, 15 MB at most. Playcut checks the file's stored duration up front. A file with no recorded duration can pass that check and fail at the provider, and a failed job is refunded.
Audio goes in on image to video only
Reference to video and text to video do not take an audio track. If the shot needs both your recording and a stack of references, it is two passes.
Five references, and one ratio list
Reference to video caps at five images. Text and reference shots take 16:9, 9:16, 1:1, 4:3 and 3:4; image to video and edit follow the source's shape instead.
Edit sources stop at 10 seconds
A longer source is a hard reject from the provider, not a truncation. Cut the segment you want before you upload.
We run 2.7, not 3.0
Alibaba has announced Wan3.0. Playcut runs Wan 2.7, so the features announced for 3.0 are not in the picker.
Same balance, other models

Wan 2.7 on one balance with three other video models

All four models below draw on one credit balance from one picker, priced in credits per second on 30 September 2026.

Model 720p 1080p Video edit Sound Refs
Wan 2.7 40 60 Yes, source 2–10 s Follows an audio track you upload, on image to video 5
HappyHorse 1.1 56 72 Yes, source up to 60 s, first 15 s edited Speaks dialogue quoted in the prompt; no audio input 9
Seedance 2.0 21 60 No Audio on every render; optional audio reference on reference mode 9
Veo 3.1 160 160 No Native audio on every clip; no audio input 3

Wan 2.7 or HappyHorse 1.1 for an edit

Both edit an existing clip by prompt, and both bill source plus output seconds. Wan is cheaper per second (40 against 56 at 720p) but takes a source of 10 seconds at most. HappyHorse accepts a source of up to 60 seconds, edits its first 15, and takes up to five reference images for a swap.

HappyHorse's edit runs on its upstream 1.0 edit model at a provisional rate. The HappyHorse 1.1 page has the details.

Your voice, or the model's

Wan follows an audio track you supply. HappyHorse 1.1 voices dialogue you quote in the prompt and takes no audio input. Seedance 2.0's reference mode can take one audio clip of up to 15 seconds as a reference for its soundtrack.

If the brief needs 4K, the Seedance 2.0 page covers the cheaper of our two 4K routes, at 108 credits a second against Veo's 240.

Image, music and voice rates sit beside these on the AI models comparison hub. For the studio side of switching models mid-project, see the AI video generator page.

Wan 2.7 on Playcut credits, not an Alibaba Cloud bill

Playcut runs Alibaba's Wan 2.7 through Alibaba's DashScope international service. You do not need an Alibaba Cloud account; renders bill against your Playcut credits and land in your Playcut workspace.

Text to video joined image to video, reference and edit on 24 July 2026, and 1080p arrived on every mode the same day. Watermarking is switched off on every Wan call we make.

Planning a talking shot around your audio

Record first, generate second. The audio decides the clip: its length sets the duration to request, and its delivery sets the performance the still has to match.

Split a long voiceover into lines of 15 seconds or less, since that is the longest clip Wan renders in one pass, and generate one shot per line. A 45-second voiceover becomes three 15-second shots: 1,800 credits at 720p, or 2,700 at 1080p.

Keep each line's still consistent if the shots cut together. The saved-prompt library helps: save the performance prompt once, then type / in an empty prompt box to reuse it on the next line.

If the cut needs a music bed under the voice, that is a separate generation on a music model. The Lyria 3 pricing page covers what a song costs in credits.

Going deeper: the Wan 2.7 API reference

Building against the API rather than the studio? Field names, file limits in full and request examples live in the Wan 2.7 API reference on docs.playcut.ai.

If you have not used the studio before, start with the guide to using Playcut, then come back to the two recipes above.

Wan 2.7 FAQ

Frequently asked questions

How much does Wan 2.7 cost on Playcut? +

40 credits a second at 720p and 60 at 1080p, on every mode. Text, image and reference shots bill output seconds, while video edit bills source plus output seconds. A 10-second 720p shot is 400 credits, and editing a 10-second source into a 10-second 720p output is 800.

Can Wan 2.7 lip-sync a still to my own voice recording? +

Yes, on image to video. Attach a WAV or MP3 of 2 to 30 seconds and up to 15 MB alongside the still, and the animation follows your audio. The clip itself runs 2 to 15 seconds, so audio longer than the clip is cut at its end.

How long can a Wan 2.7 video edit be? +

The source must be 2 to 10 seconds, MP4 or MOV, and 100 MB or less. Longer sources are rejected, not trimmed. The output runs 2 to 10 seconds and defaults to 5, whatever the source length.

Does Wan 2.7 render 4K? +

No. Wan 2.7 renders at 720p or 1080p. For 4K, Seedance 2.0 and Veo 3.1 are the two video models we run that reach it.

Is this Wan 3.0? +

No. Playcut runs Wan 2.7. Alibaba has announced Wan3.0, but we do not run it, so its features are not in the picker or on this page.

Does Playcut rewrite my Wan 2.7 prompt? +

Playcut does not, but Alibaba does. Every Wan request goes out with Alibaba's prompt expansion switched on, which rewrites and expands the prompt before rendering, so the output will not follow your wording literally.

Ready to make something cinematic?

Pick a plan. Every plan bills at checkout — cancel anytime.