Painted fox
Image to video · 10s · 720pThe illustrated fox trots across the snowy meadow, leaves paw prints, then stops and looks back. Keep the painted style of the image. Soft wind, crunching snow.
Start image
Alibaba’s newest Wan model. Start from a photo, keep a character identical with up to ten reference images, and render a single take up to thirty seconds at 1080p with sound.

5 credits
for 5s at 480p (≈ $0.95)
88s
across 2,516 finished clips, last 60 days
79/100
#3 of 9 engines, blind benchmark
Live figures from the GoCrazy Video Index, refreshed hourly. Dollar amounts use the best plan rate.
Every clip below was generated on GoCrazyAI with the exact prompt and settings shown. Hover or tap to play.
The illustrated fox trots across the snowy meadow, leaves paw prints, then stops and looks back. Keep the painted style of the image. Soft wind, crunching snow.
Start imageThe dog in the photo tilts its head, ears perk up, then it trots toward the camera wagging its tail. Low camera at the dog’s eye level. Paws on grass, a happy bark.
Start imageFishing boats rock gently in a small harbor at sunset, lights turning on along the quay one by one. Slow pan left to right. Water lapping, distant voices.
Start imageWan 3.0 Prime is the Prime version of Alibaba’s Wan 3.0 video model. It is built around a start image: your photo becomes the first frame, and the prompt tells it what moves. Add a last-frame image and it resolves the shot on that picture.
Its standout is reference-to-video. Attach up to ten images of the same person, product or place (plus up to five clips and five audio files) and Wan keeps that subject consistent while it invents the motion. That makes it the engine to reach for when a character must look the same across a series of clips.
It renders at 480p, 720p or 1080p, holds a take for up to thirty seconds, and generates audio in the same pass. It is also the default engine in the Extend tab, because it continues a clip from its last frame at full 1080p.
Credits, from the live price list. You always see the exact cost before you generate.
| Resolution | 5s | 10s | 15s | 30s |
|---|---|---|---|---|
| 480p | 5≈ $0.95 | 10≈ $1.90 | 15≈ $2.85 | 30≈ $5.70 |
| 720p | 10≈ $1.90 | 20≈ $3.80 | 30≈ $5.70 | 59≈ $11.21 |
| 1080p | 20≈ $3.80 | 40≈ $7.60 | 59≈ $11.21 | 118≈ $22.42 |
Dollar figures use the best plan rate. See plans and credit packs.
Ten reference images is the most of any engine in the studio. Give it several angles of one person and the character stays recognisably the same across a whole series.
Full HD and a thirty-second take in one generation, so a finished clip needs no upscale and no stitching.
At 480p it is one of the cheapest ways to draft a long take. Step up to 1080p only for the version you keep.
As the default Extend engine, Wan picks up from a clip’s last frame with matching light and motion, including clips made on other engines.
Feed it the packshots you already have and the label, shape and colour stay correct while the camera moves around the product.
Footsteps, room tone and short spoken lines come out synced to the picture, with no separate audio step.
Wan 3.0 Prime is preselected when you arrive from this page. Add a start image, or leave it empty to generate from text, or attach references.
Describe the subject, the action, the camera and the light in one or two sentences. Start from a tested prompt below.
Choose duration, resolution and aspect ratio. The credit cost updates as you change them, and failed renders are refunded.
Six prompts written for how Wan 3.0 Prime reads direction. The full guide has 30 more, sorted by use case.
The woman from the reference images walks into a sunlit cafe, looks around, and sits by the window with a book. The camera follows at shoulder height. Soft cafe chatter, a cup set down on a saucer.
Attach 3 to 6 photos of the same person from different angles.
The man in the photo turns his head toward the window as the light shifts, then smiles slightly. Gentle breeze moves the curtain. Camera locked off. Quiet room tone and distant birds.
The camera orbits slowly around the product from the reference images on a stone pedestal, soft morning light raking across its surface. Clean background. Low ambient hum.
Render at 1080p for a store-ready clip.
A woman in a yellow raincoat walks down a neon-lit street in light rain, first passing a noodle stall, then crossing at the lights, then stepping into a doorway. Tracking shot from the side. Rain, sizzling pans, traffic.
Starting on a misty forest floor, the camera rises slowly through the trees and breaks above the canopy, ending on the mountain view in the last-frame image. Birdsong fading into wind.
The illustrated fox in the image trots across a snowy meadow, leaving small paw prints, then stops and looks back over its shoulder. Keep the painted style. Soft wind and crunching snow.
Same studio, same credits. Switch engines per clip; here is when another one fits better.
| Engine | Longest clip | Audio | Credits / 5s | Median render |
|---|---|---|---|---|
| Seedance 2.5 | 30s | Native | 30 | 385s |
| Seedance 2.5 Turbo | 30s | Native | 16 | — |
| Wan 3.0 Prime | 30s | Native | 10 | 88s |
| MiniMax H3 | 15s | Native | 4 | 108s |
| Seedance 2.0 Mini | 15s | Native | 10 | 227s |
| Seedance 2.0 Fast | 15s | Native | 15 | 213s |
| Seedance 2.0 | 15s | Native | 20 | 242s |
| Wan 2.7 | 15s | Native | 17 | 114s |
| HappyHorse 1.1 | 15s | Native | 19 | 110s |
Credits for a 5-second clip at 720p (or the engine's nearest tier). Full method on the Video Index.
Wan 3.0 Prime is the Prime version of Alibaba’s Wan 3.0 video model. It turns a photo into video up to 30 seconds long at up to 1080p, with native audio, and can follow up to ten reference images to keep a subject consistent.
It needs at least one image: a start frame or a reference. For prompt-only video, Seedance 2.5 and HappyHorse 1.1 generate from text alone.
Up to 10 images, plus up to 5 video clips and 5 audio files. Reference mode is billed at a higher rate than plain image-to-video; the generator shows the exact cost first.
It is billed in credits by length and resolution. The table on this page comes from the live price list, and failed renders are refunded automatically.
480p, 720p or 1080p, and 5, 10, 15 or 30 seconds per generation. Extend a clip afterwards by another 5 to 30 seconds.
Yes. Audio is generated with the video and synced to it. Describe two or three sounds in the prompt to steer it.
Yes, as the start frame, the last frame or references. Uploads are checked automatically. GoCrazyAI is for adults (18+) only, and anyone in your photos must be an adult who has agreed to appear.
Opens AI Video Pro with Wan 3.0 Prime selected. Paste a prompt from this page and adjust the length before you generate.
Last updated: