Generate Images and Video Clips With Wan 2.7
Last verified: November 3, 2026
Wan2.7-Video can turn a single prompt into a full multi-shot storyboard . Its sibling Wan2.7-Image is built for precise color control , so the motion and the key art for one campaign can come from the same family. Wan 2.7 is Alibaba's combined image and video release. Alibaba announced Wan2.7-Image and the higher-tier Wan2.7-Image-Pro in Hangzhou on April 1, 2026 . Wan2.7-Video came out a few days later as four models: Wan2.7-t2v for text-to-video, Wan2.7-i2v for image-to-video, Wan2.7-r2v for reference-to-video and Wan2.7-videoedit for video editing . The generator on this page puts the family's image and video modes side by side, along with a Pro tier for stills.
On the video side, the biggest change is in image-to-video. Alibaba's earlier image-to-video API (Wan 2.6 and before) could only animate from a first frame. Wan2.7-i2v can also build a clip from a first frame and a last frame , and Alibaba lists it with multi-shot narrative support . On the image side, Wan2.7-Image aims to replace the generic AI look with personalized results and is a big step forward in text rendering. Wan2.7-Image-Pro adds more stable composition, a sharper reading of prompts and high-definition 4K output .
Explore Wan AI's Models
Inputs, Outputs and Limits
The upload rules, output sizes and model limits behind every image and clip you make here.
Still image sizes
1K or 2K; Wan 2.7 Pro Text to image adds 4K
Images per Image to image run
1 to 9 (JPG, JPEG, PNG or WEBP, up to 10 MB, at least 300×300 px)
Thinking Mode
On/off switch in Wan 2.7 and Wan 2.7 Pro Text to image, off by default
Video frame rate and file
30 fps MP4 from Wan2.7-i2v and Wan2.7-r2v
Reference cast limit
5 reference images and videos combined in Wan2.7-r2v
Reference to video uploads
Reference images, reference videos (MP4 or MOV, up to 50 MB and 45 s) and an optional First Image Frame
Check These Before You Hit Generate
Six model-specific checks that prevent mismatched frames, mixed-up references and muddy layouts.
Number your images in Image to image prompts
Wan2.7-Image reads multi-image instructions by image number. Alibaba's own example asks it to move graffiti from image 2 onto the car in image 1 . Add files to Add Image in the order your Prompt names them, and say what each one contributes: a pattern, a product, a backdrop.
Turn on Thinking Mode for crowded layouts
In Text to image (Wan 2.7 and Wan 2.7 Pro), Thinking Mode makes the model reason more before it renders. That improves image quality but takes longer . Turn it on for scenes with several subjects, infographics or tight spatial arrangements, and leave it off while you sketch out ideas.
Write shot-by-shot beats for storyboard clips
Wan2.7-Video can return a whole storyboard from one prompt , so give each shot its framing, action and rough timing in the Prompt. Make sure the shots fit the Video Duration you pick. Four beats crammed into a 3-second clip leave none of them time to register.
Pair a First Frame and Last Frame that can connect
First to last frame needs both uploads, and the model fills in everything between them. Keep the same subject, lens feel and lighting direction in both stills, and use the Prompt to name the move (a push-in, a turn, a change of time of day). The more the two images have in common, the less the transition has to invent.
Label references as Image 1 and Video 1
In Reference to video, the model tells uploads apart by labels such as Image 1 or Video 1. Images and videos are numbered separately, and with a single reference, 'the reference image' is enough . Use the same labels in your Prompt so each character lands in the right role.
Use Negative Prompt to keep clutter out of clips
Text to video and Image to video both have a Negative Prompt field. List what must not appear there, such as on-screen captions, watermarks, extra fingers or warped logos, instead of filling the main Prompt with 'no' clauses that compete with your scene description.
Wan 2.7 or Kling 3.0: Match the Generator to the Shot
Wan 2.7 and Kling 3.0 are both in the same model list. The first six rows compare the options each generator shows on screen: modes, Resolution, Video Duration, Aspect Ratio, sound and prompt length. The last two rows give what Alibaba and Kuaishou say about the models themselves. Use the table to match each job to the generator that fits it.
| Feature/Spec | Wan 2.7 | Kling 3.0 |
|---|---|---|
| Modes in the model list | Wan 2.7: Text to image, Image to image, Text to video, Image to video, First to last frame, Reference to video. Wan 2.7 Pro: Text to image, Image to image | Kling 3.0: Text to video, Image to video, First to last frame, Motion control. Kling 3.0 Turbo: Text to video, Image to video |
| Video Resolution options | 720P or 1080P in Text to video, Image to video, First to last frame and Reference to video | 720P, 1080P or 4K (Kling 3.0 Text to video, Image to video, First to last frame); 720P or 1080P (Kling 3.0 Turbo); 720p or 1080p via Mode (Kling 3.0 Motion control) |
| Video Duration options | 3–15 s (Text to video, Image to video, First to last frame); 3–10 s (Reference to video) | 3–15 s (Kling 3.0 Text to video, Image to video, First to last frame; Kling 3.0 Turbo Text to video, Image to video) |
| Aspect Ratio options in Text to video | 1:1, 3:4, 4:3, 9:16, 16:9 (default 1:1) | Kling 3.0: 1:1, 9:16, 16:9 (default 1:1) |
| Sound generated with the clip | The Wan 2.7 generator here has no such setting | Generate Audio switch, on by default (Kling 3.0 Text to video, Image to video, First to last frame) |
| Prompt length | Up to 2,048 characters in every mode | Up to 2,500 characters (Kling 3.0 Image to video, Motion control); up to 2,048 in the other Kling 3.0 and Kling 3.0 Turbo modes |
| Multi-shot storytelling | Wan2.7-Video: a single prompt can yield a full storyboard | Video 3.0 follows multi-scene, multi-shot instructions and adjusts camera angles and shots to match |
| Release date | Wan2.7-Image: April 1, 2026 ; Wan2.7-Video: a few days later | Kling AI 3.0 models: February 5, 2026 |
Matching Each Generator to the Job
Finished masters: 4K and built-in sound
Sometimes the clip itself is the deliverable: a large-screen loop, a spot that needs a 4K master, or a social ad that has to play with sound on. For that, Kling 3.0's Text to video, Image to video and First to last frame modes offer a 4K step above 1080P and a Generate Audio switch that adds sound in the same run.
Wan 2.7's video modes work at 720P or 1080P, which covers web and social delivery. They also sit next to its image modes, which helps when one campaign needs both stills and clips.
Reusing a cast vs. working from a performance clip
Wan 2.7's Reference to video mode is built around identity. You supply stills or short clips of a person, animal or object, and Wan2.7-r2v keeps their appearance as it builds new scenes around them . Kling 3.0's Motion control takes one character image plus a video of up to 30 seconds, and its on-screen help asks you to keep the character in frame and avoid cuts or camera movement.
Choose by what you start with: a cast you want to place in new scenes, or recorded footage you want to build from.
Frame shapes, prompt room and where they match
In Text to video, Wan 2.7 offers 3:4 and 4:3 alongside square, vertical and widescreen, so feed posts and slide-deck clips need no crop. Kling 3.0's Image to video and Motion control accept longer prompts, which helps with dense shot lists. Clip length is the same in the video modes both run, and both creators describe multi-shot storytelling, so neither duration nor shot structure should decide it on its own.
Decide Between Wan 2.7 and Kling 3.0 by What You Deliver
Use this quick guidance to pick the best option for your workflow.
How to choose: Choose Wan 2.7 when the job needs images as well as video. It has Text to image and Image to image modes, and Wan 2.7 Pro Text to image goes up to 4K. It's also the pick for casting characters from reference images and videos in Reference to video, and for Text to video clips in a 3:4 or 4:3 frame. Choose Kling 3.0 when a clip needs a 4K master or sound generated with it (Text to video, Image to video and First to last frame). It's also the pick for building from a character image plus a video in Motion control, and for Image to video prompts longer than 2,048 characters, up to 2,500. The two are the same on 720P and 1080P output and on 3–15 second clips in the video modes they share, and both creators describe multi-shot storytelling.
From Idea to Finished Image or Clip
Five steps take you from choosing a mode to a generated result.
Choose a Wan 2.7 or Wan 2.7 Pro mode
Open the model list and pick the mode that fits the job. Wan 2.7 offers Text to image, Image to image, Text to video, Image to video, First to last frame and Reference to video. Wan 2.7 Pro offers Text to image and Image to image. Each mode shows its own uploads and settings.
Add files in the upload fields
Text to image and Text to video need no uploads. For Image to image, add one to nine pictures under Add Image. For Image to video, add one Image. For First to last frame, use Upload first frame and Upload last frame. For Reference to video, add at least one Reference Image or Reference Video , plus an optional First Image Frame. Images must be JPG, JPEG, PNG or WEBP, at least 300×300 pixels and no larger than 10 MB.
Write the Prompt and try Enhance Prompt
Describe the subject, action, camera and look in the Prompt box, up to 2,048 characters. Click Enhance Prompt under the box to get fuller versions of a rough idea. Your text only changes if you pick one of them.
Set Aspect Ratio, Resolution and Video Duration
Image modes have Aspect Ratio and Resolution Quality, and Text to image also has the Thinking Mode switch. Video modes have Video Duration and Resolution. Text to video and Reference to video add Aspect Ratio, and Text to video and Image to video add a Negative Prompt field. Every mode has an optional Seed field.
Press Generate
Click Generate to make one image or one clip. Then adjust the Prompt, settings or Seed and run again to try alternatives.