WAN 2.5 (Image-to-Video)

The current Masonry route requires a text prompt, one image, and a dimension choice. It also exposes a negative prompt, seed, and the video command's duration option. The public parameter view does not expose an audio upload, lip-sync control, multiple reference images, a first/last-frame pair, or a motion brush.

The endpoint accepts five- or ten-second durations. Start with five seconds when testing a new product, motion pattern, or prompt because the first question is whether the returned frames preserve the approved asset. A longer clip creates more temporal surface area to inspect and should have a placement reason.

Masonry exposes 854×480, 1280×720, and 1920×1080 dimensions, corresponding to 480p, 720p, and 1080p. The underlying fal endpoint documents the same three resolution tiers. Resolution does not repair geometry drift, so review identity before paying for the final tier.

Do not assume exact preservation. In one retained Masonry test, the fictional stand's main structure remained stable across five one-second samples, but the camera pushed in substantially more than requested. Different products, prompts, seeds, and service versions can behave differently. Compare every commercially relevant frame with approved source evidence.

One 1280×720, five-second job used an approved 16:9 image of a fictional graphite folding stand and requested a four-percent push-in plus one light sweep. Across five sampled seconds, no new prop, text, cut, fold, or background replacement appeared, and the visible product structure remained recognizable. The stronger-than-requested zoom means the output is useful identity evidence but not a strict camera-direction pass. One output is not a success rate.

Not through the current public CLI parameter view for this model. The underlying fal endpoint documents an optional audio URL, but Masonry's exposed WAN 2.5 inputs do not currently include it. This page therefore does not instruct Masonry users to upload audio or promise a lip-sync workflow.

The documented fal image-to-video schema describes optional supplied audio rather than a native-audio control, and Masonry does not expose that upload here. The retained test file contained an AAC track with near-silent measured levels, which should not be treated as useful generated sound. Plan narration, music, captions, and compliance review as a separate delivery step.

Masonry's dimension control accepts 16:9, 1:1, and 9:16 aliases as well as the listed pixel dimensions. Compose the approved source at the placement's intended ratio and choose the matching dimension. Then verify the returned file instead of relying on an assumed follow-input rule.

No self-hosting workflow or downloadable weights are exposed by this Masonry route; it calls the hosted fal image-to-video endpoint. Do not treat the WAN 2.5 API name as evidence that weights or a particular license are available. Teams that require self-hosting should evaluate a separately documented open-weight model and its license.

One image can support visible surface appearance and restrained motion, but it cannot establish hidden geometry or how a mechanism operates. For a documented fold, opening, or adjustment, use approved first and last states with a route that accepts both, then inspect the intermediate frames. Record the real product or use verified 3D when the motion itself is evidence.

Check the first frame, last frame, and regular samples between them. Compare silhouette, proportions, part count, labels, color, material, reflections, background, crop, camera behavior, and every claimed motion with the approved record. Mark each item pass, hold, or reject. Keep generated offer text and pricing out of the master so those fields remain deterministic.

No. Render price, promotion, SKU, CTA, captions, disclaimers, and legal copy in a deterministic editor or storefront layer after the clean motion passes. This prevents a regeneration from changing customer-facing commerce state and lets a merchant update an offer without recreating the video.

Use another method when the clip must prove a mechanism, show an unseen side accurately, preserve tiny regulated label text through aggressive motion, deliver an exact first-to-last transition, or maintain a repeatable camera path. Those jobs need stronger source coverage, first/last-frame control, verified 3D, or conventional capture.

Measure cost per accepted clip first: generation cost plus review and rework divided by outputs that pass the channel gate. After release, join eligible exposure to play behavior, selected variant, add-to-cart, checkout, completed order, fulfillment, returns, and contribution. A valid experiment is required before claiming incremental revenue; views and assisted conversions alone are directional.

Use it for a bounded still-to-motion draft after the exact SKU image and allowed motion are approved. The source-accurate PDP workflow then adds temporal inspection, deterministic overlays and poster imagery, exact variant mapping, accessible playback, page-performance checks, channel validation, and completed-order measurement. WAN 2.5 is one generation step, not the whole release system.

WAN 2.5 (Image-to-Video) is an AI video generation model from WAN Video, available inside Masonry, the AI creative agent teams use to produce marketing, product, and brand videos.

Open a Masonry canvas, pick WAN 2.5 (Image-to-Video) from the model selector, and describe the video you need: a product shot, an ad creative, a social post. Masonry generates it, then you refine, edit, and combine WAN 2.5 (Image-to-Video) with other models in one workspace.

Yes, you can start generating videos with WAN 2.5 (Image-to-Video) on Masonry's free tier, then scale up with higher limits and priority processing as your team grows.

Use one approved product image at the intended delivery aspect. Ask for one restrained camera or lighting behavior, explicitly freeze geometry and object count, and list the expensive failure modes in the negative prompt. Start with a five-second 720p review. Sample frames across time before deciding whether a 1080p rerun is worth the cost. Add price, offer, CTA, captions, and legal copy after generation so they remain exact and editable. See the prompt gallery on this page for real WAN 2.5 (Image-to-Video) prompts you can copy and adapt.

WAN 2.5 (Image-to-Video) is built by WAN Video. Inside Masonry it runs alongside 50+ image and video models, so your team can pick the right one for each brief without switching tools.

Yes, the prompt gallery on this page shows real videos teams have generated with WAN 2.5 (Image-to-Video) in Masonry, each paired with the exact prompt you can copy and adapt for your own brand.