Seedance 2.5 vs Wan 3.0 vs Vidu Q4: one prompt, three video models
Pick Seedance 2.5 for the final cut with many references, Wan 3.0 when you need a first or last frame or the lowest price per second ($0.055/s at 720P), and Vidu Q4 to turn a few photos into a short 720P clip at $0.065/s. All three run on the same VidBudget API key and request format; only model and a few limits change. Below, the same official case on each, then the specs and prices side by side.
Short answer
- Seedance 2.5 (ByteDance) takes the most references (up to 30 images, 10 videos and 10 audio clips), 4 to 30 seconds up to 1080P, and follows long, timed storyboards most closely.
- Wan 3.0 (Alibaba) is the only one here with a first frame or a first and last frame (
"image_role": "first_frame"), runs 2 to 30 seconds at 480P, 720P or 1080P, and bills a reference video's seconds at the output rate. - Vidu Q4 (Shengshu) makes a video from 1 to 10 reference images plus up to 3 audio clips, 3 to 16 seconds at 720P; it needs at least one image and takes no reference video.
The same case on all three
The case is the "one-click video" example of ByteDance's Seedance 2.5 prompt guide: eight photos of a puppy in different outfits and one prompt asking for a hand-drawn, doodle-style coffee-shop vlog with playful sound, keeping the photos close to the originals. We sent the guide's Chinese prompt and its photos unchanged to each model at 720P, 9:16, 15 seconds.
What we saw: Seedance 2.5 opened on a hand-drawn frame, then cut through the photos with light motion, close to the originals. Wan 3.0 also opened in a drawn style, then restaged several photos with new props (a teddy bear, a latte) and gave the dog the most movement. Vidu Q4 kept each photo closest to the original, like animated live photos, without the drawn style. This is one case, rendered once on each model; no single one wins for every job.
The prompt as sent
Turn all the images into one video, in any order: a coffee-shop vlog in a hand-drawn animated doodle cut-out style, following a puppy in different cute outfits as it poses for photos at a coffee shop on a fun day out. Add playful, internet-style audio or background music. The images can move slightly, like live photos, but do not change the originals; keep them highly consistent with the source images.
Specs side by side
| Seedance 2.5 | Wan 3.0 | Vidu Q4 | |
|---|---|---|---|
| Maker | ByteDance | Alibaba | Shengshu |
| Model ID | seedance-2.5 | wan-3.0 | vidu-q4 |
| Duration | 4–30 s | 2–30 s | 3–16 s |
| Resolution | 480P / 720P | 480P / 720P / 1080P | 720P |
| Aspect ratios | 16:9, 9:16, 1:1, 4:3, 3:4 | 16:9, 9:16, 1:1, 4:3, 3:4 | 16:9, 9:16, 1:1, 4:3, 3:4 |
| Reference images / videos / audio | 30 / 10 / 10 | 10 / 5 / 5 | 10 / 0 / 3 |
| Text only | Yes | Yes | No, 1 image at least |
| First / last frame | No | Yes | No |
| Sound in the output | Yes | Yes | Yes |
| Content review | ByteDance | Alibaba (strict on real faces) | Shengshu (refuses public figures) |
Sound is generated with the picture on all three; "audio": false does not guarantee silence. Render times vary with the queue: in our own test renders on 2026-11-20, Wan 3.0 took 1 to 9 minutes and Vidu Q4 2.5 to 5 minutes; Seedance 2.5 usually takes several minutes.
Price per second and per clip
| Model · resolution | Per second | 10-second clip | Official reference |
|---|---|---|---|
| Seedance 2.5 · 480P | $0.10 | $1.00 | $0.103/s |
| Seedance 2.5 · 720P | $0.158 | $1.58 | $0.231/s |
| Wan 3.0 · 480P | $0.03 | $0.30 | $0.035/s |
| Wan 3.0 · 720P | $0.055 | $0.55 | $0.07/s |
| Wan 3.0 · 1080P | $0.11 | $1.10 | $0.14/s |
| Vidu Q4 · 720P | $0.065 | $0.65 | $0.067/s |
Prices are read from the live price list. With a reference video, Seedance 2.5 at 720P and Wan 3.0 bill the reference seconds too; reference images and audio are free on all three. Failed tasks are refunded in full. Official references: BytePlus ModelArk for Seedance, Alibaba Cloud Model Studio for Wan (its current 30% off price), and Vidu's own API for Vidu. Every price, with the top-up bonus, is on the pricing page.
One request format
All three go to POST /v1/videos/generations with the same fields. To switch, change model and keep within that model's limits:
{"model": "wan-3.0", "prompt": "A cat running on the grass", "duration": 5, "resolution": "720P",
"aspect_ratio": "16:9", "images": ["https://.../first.png"], "image_role": "first_frame"}
{"model": "vidu-q4", "prompt": "The puppy trots through a sunny coffee shop", "duration": 8,
"resolution": "720P", "aspect_ratio": "9:16", "images": ["https://.../puppy.jpg"]}
Each model page has the full limits and an online test: Seedance 2.5, Wan 3.0, Vidu Q4.
FAQ
Is Wan 3.0 better than Seedance 2.5?
It depends on the job. Seedance 2.5 follows long, timed storyboards with many references most closely; Wan 3.0 costs less per second and is the only one of the two with a first or last frame. Run the same prompt on both: the request format is the same.
Which one costs the least?
Per second at 720P, Wan 3.0 ($0.055/s). Vidu Q4 is 720P only at $0.065/s, and Seedance 2.5 Flash is the low-cost Seedance.
Can I use photos of real people?
Each maker reviews its own model. Wan may refuse a reference image that shows a real face, Vidu refuses public figures, and Seedance may refuse real faces too. Sexual content of a real person without consent is never allowed, and sexual content involving minors is refused at submission.
Does Vidu Q4 do text-to-video?
Not here: Vidu Q4 makes a video from reference images, so send at least one image. For text only, use Seedance 2.5 or Wan 3.0.