WAN 2.2 Image-to-Video
Animate a photo with WAN 2.2 image-to-video: 480p/720p, 5–8s clips, 30–60 credits/s, negative prompt and last-frame control. See how it compares with Spicy, WAN 2.5 and WAN 2.6.
WAN 2.2 Image-to-Video is Alibaba's standard image-to-video checkpoint built on a mixture-of-experts denoising architecture. Give it one still image and a motion prompt, and it returns a 5- or 8-second clip at 480p or 720p. Unlike the Spicy variant, the standard model exposes a negative prompt, a last-frame image and a seed, which makes it the more controllable choice for product shots, character continuity and transitions. On Conut.ai it costs 30 credits per second at 480p and 60 credits per second at 720p.
- Resolution: 480p / 720p
- Duration: 5s / 8s
- Price: 30–60 credits/s
- Controls: Negative prompt + last frame
How to use WAN 2.2 Image-to-Video
This is the workhorse image animation model. Use the controls in the order they affect the result: frames first, prompt second, price last.
- Upload the first frame Choose a photo with a clear subject and enough surrounding space for the described motion. The model preserves the source composition.
- Add a last frame when the clip must land somewhere A second image turns open-ended animation into a transition. Keep scale, lighting and framing close between the two frames or the motion bends awkwardly.
- Describe motion, not appearance The image already fixes appearance. Spend the prompt on what happens: a slow dolly in, the subject turning, steam rising from a cup.
- Set a negative prompt for known artefacts Warped hands, extra fingers and garbled text are the usual failures. Listing them is cheaper than rerolling until they disappear.
- Choose length and resolution Test prompts at 5 seconds and 480p, then move to 8 seconds and 720p for the final run. A 5-second 720p clip costs 300 credits.
Settings reference
Standard WAN 2.2 Image-to-Video is the control-oriented silent checkpoint: first frame, optional last frame, negative prompt and seed.
| Setting | Value | Notes |
|---|---|---|
| First frame | 1 image, required | The clip starts from this frame and keeps its composition and identity. |
| Prompt | Required | Describe scene mood, motion and camera cues. |
| Last-frame image | Optional | Supply an end state and the model plans the motion to land on it. |
| Negative prompt | Optional | Name artefacts to avoid, such as extra limbs, warped faces or text. |
| Duration | 5s / 8s | Default is 5 seconds. |
| Resolution | 480p / 720p | 30 credits per second at 480p, 60 at 720p. 720p is the default. |
| Seed | Optional | Reuse a seed to keep the motion while you change the prompt. |
What WAN 2.2 image-to-video does well
The standard WAN 2.2 checkpoint is tuned for coherence rather than maximum motion. It keeps the subject on model, follows detailed scene descriptions and produces a camera move that feels deliberate instead of improvised. The model understands lighting, color, composition and basic camera language, so prompts written like a shot description usually land closer than a list of adjectives.
It is the right default when the source image is already the shot you want and you need the animation to preserve it. Portraits keep their likeness, products keep their shape and labels, and landscapes keep their horizon. That stability is also why the standard model costs more than Spicy per second: you are paying for control, not for extra movement.
- Best for controlled cinematic motion, product shots and character continuity
- Understands lighting, lens and camera direction in the prompt
- Keeps composition and identity closer to the source image than Spicy
- Works at 480p for drafts and 720p for final delivery
WAN 2.2 vs Spicy, WAN 2.5 and WAN 2.6
All four models animate a still image, but they optimise for different things. Spicy pushes motion and contrast; WAN 2.5 adds audio and 1080p; WAN 2.6 extends the clip to 15 seconds. Pick by the deliverable, not by version number.
- Choose standard WAN 2.2 when you need a negative prompt or a defined last frame.
- Choose Spicy when the shot feels too flat and you want more movement for less per second.
- Choose WAN 2.5 when the subject speaks or needs synchronized audio.
- Choose WAN 2.6 when the shot must run longer than 8 seconds or reach 1080p.
| Model | Resolution | Duration | Credits | Audio | Standout control |
|---|---|---|---|---|---|
| WAN 2.2 I2V | 480p / 720p | 5s, 8s | 30–60 /s | No | Negative prompt, last frame |
| WAN 2.2 Spicy | 480p / 720p | 5s, 8s | 20–40 /s | No | Stronger motion and contrast |
| WAN 2.5 I2V | 480p / 720p / 1080p | 5s, 10s | 20–40 /s | Yes, one-pass A/V sync | Dialogue, lip sync, audio upload |
| WAN 2.6 I2V | 720p / 1080p | 5s, 10s, 15s | 30–50 /s | Yes | 15-second clips |
Negative prompt and last-frame control
The negative prompt is a list of things the model should avoid. On this endpoint it is the cleanest way to suppress common failures: extra fingers, warped hands, text-like artifacts, watermarks, split faces and duplicate limbs. Keep the list short and concrete. A long negative prompt that repeats the same idea in five different ways usually does less than four specific terms.
The last-frame image is the other control that Spicy does not expose. Upload a start image and an end image, and the model animates the gap between them. This is useful for match cuts, product reveals, before-and-after shots and any clip that has to land on a specific composition. The start and end frames should share the same subject, framing and lighting, otherwise the model has to invent a transition and the middle of the clip can warp.
- Negative prompt: name the artifact, not the whole bad scene
- Last frame: keep the same subject, crop and light direction as the first frame
- Use a seed when comparing prompt variants so the starting noise stays fixed
- For a simple one-image animation, skip the last frame and let the model move freely
How to prompt WAN 2.2 for stable motion
Write the prompt as subject, action, camera, light and style. One dominant action per clip is enough; a turn, a step, a head lift or a wingbeat reads better than four actions packed into five seconds. Put the camera move in its own clause so the model does not blend it into the subject's motion.
- Name one action: turns toward camera, walks two steps, lifts its head, unfolds its wings.
- Name one camera move: slow dolly-in, 30-degree orbit, low tracking shot, tilt up.
- Name the light: backlit, warm rim light, overcast softbox, neon reflections.
- Keep the source image clean: one subject, clear silhouette, no text or busy collage.
- Draft at 480p, then re-run the winning prompt at 720p for the final clip.
Limitations and common failure cases
WAN 2.2 image-to-video tops out at 720p and eight seconds, and it produces silent video. Faces, hands, text and fine repeating patterns are the usual failure points, especially when they move quickly or sit near the edge of the frame. The model also inherits the quality ceiling of the source image: a soft, noisy or cluttered input stays soft no matter how detailed the prompt is.
- Need 1080p or audio? Use WAN 2.5 or WAN 2.6 image-to-video.
- Need a clip longer than 8 seconds? Use WAN 2.6 or join multiple clips in an editor.
- Keep hands and faces small in frame or slow them down when they must stay readable.
- Avoid fast lateral motion across a detailed background; it is the quickest way to get warping.
- If the shot needs more energy than control, switch to WAN 2.2 Spicy.
Prompt examples for WAN 2.2 Image-to-Video
- Cinematic corridor walk: A person walking through a corridor of shattered clocks and frozen time fragments, motion blur trails, glowing fragments suspended midair, slow dolly-in, surreal lighting, shallow depth of field. One action, one camera move, one light idea. A good first test for motion stability.
- Product turntable: A matte black wireless speaker on a dark stone pedestal, camera orbits 30 degrees around the product, soft studio highlights sliding across the grille, subtle dust particles in the air, luxury commercial lighting. Rigid objects hide the model's weakest areas and make pricing tests easy to compare.
- Portrait with a small action: A woman in a dark green coat turns her head slowly toward the window, rain running down the glass, cool daylight on one side of her face, warm lamp light on the other, slow push-in, cinematic portrait. Small head movement is the safest way to keep a face stable.
- Landscape reveal: Morning fog rolling over a pine forest ridge, the first sunlight breaking through the trees, slow aerial push-in, volumetric light rays, cold green and amber palette, natural documentary style. Landscapes tolerate large camera motion without a subject to warp.
- Anime rooftop: Anime heroine standing on a rooftop at sunset, coat flapping in the wind, petals drifting past, camera tilts up toward the skyline, hand-painted background, warm rim light, film still. Stylized art hides small artifacts that would be obvious in live action.
- Last-frame transition: Start with a closed red flower in a dark studio; the petals open as the camera slowly pushes in, water droplets falling, the light shifts from cool blue to warm gold, macro cinematography. Upload a matching end frame to land on a specific composition.
Pro tips for WAN 2.2 Image-to-Video
The last-frame control is the strongest lever on this page; most other settings are about cost and reproducibility.
- A last frame beats a longer prompt When the endpoint matters, an end image communicates the target far more precisely than several sentences about where the subject should stop.
- Lock the seed before comparing prompts Changing the prompt and the seed at the same time tells you nothing. Keep the seed fixed so the comparison isolates the wording.
- 480p for drafts, 720p for delivery Half-price drafts are the fastest way to explore motion ideas. Only the selected run needs to be rerun at 720p.
- Name artefacts in the negative prompt Short lists work better than long ones: extra fingers, distorted face, text, watermark. Too many terms start excluding legitimate detail.
- Use Spicy only when you want more energy Standard WAN 2.2 is the safer default for product and portrait motion. Spicy is the deliberate choice for stylized, high-amplitude movement.
Frequently asked questions
What is WAN 2.2 image-to-video?
WAN 2.2 Image-to-Video is Alibaba's standard image animation model. It turns one still image plus a motion prompt into a 5- or 8-second clip at 480p or 720p, with optional negative prompt, last frame and seed controls.
How much does WAN 2.2 image-to-video cost on Conut.ai?
480p costs 30 credits per second and 720p costs 60 credits per second. A 5-second 480p clip is 150 credits, a 5-second 720p clip is 300 credits, and an 8-second 720p clip is 480 credits.
What is the difference between WAN 2.2 and WAN 2.2 Spicy?
Both use the WAN 2.2 architecture. Spicy is tuned for larger motion, stronger contrast and richer color, and it costs less per second. The standard model is more conservative and exposes a negative prompt and last-frame image, which gives you more control over the result.
Does WAN 2.2 image-to-video generate audio?
No. This endpoint returns silent video. For dialogue, voiceover or synchronized sound, use WAN 2.5 or WAN 2.6 image-to-video.
Can I control the end of the clip with a last-frame image?
Yes. Upload a start image and an end image and the model animates the transition between them. Keep both frames consistent in subject, framing and lighting for the cleanest result.
What should I put in the negative prompt?
Name the specific artifacts you want to avoid: extra fingers, warped hands, text, watermark, split face, duplicate limbs. Keep it short and concrete rather than describing an entire bad scene.
Can WAN 2.2 image-to-video output 1080p?
No. This endpoint supports 480p and 720p. Use WAN 2.5 or WAN 2.6 image-to-video when you need 1080p.
What images work best as input?
Use a sharp image with one clear subject, a readable silhouette and a simple background. Avoid text, logos, heavy motion blur and busy collages. The model preserves what it is given, so a clean source image is the easiest quality upgrade.
Compare with other image-to-video models
- WAN 2.2 Spicy
- WAN 2.5 Image-to-Video
- WAN 2.6 Image-to-Video
- Kling Image-to-Video
- Seedance 2.5 Image-to-Video
- Browse all AI video models
Specifications
- Provider: Wavespeed
- Model type: Image to video
- Credit pricing: 480p: 30 credits · 720p: 60 credits
- Capabilities: popular, cinematic, realistic, multi-shot
About Conut.ai
Conut.ai is an AI image, video and audio generation platform with 50+ creative tools. It supports text-to-image, image-to-video, face swap, motion control, relighting, inpainting and video effects through models including Nano Banana Pro, Seedream, GPT Image, Sora 2, Kling, WAN and Seedance. Start free — no credit card required.