Wan 3.0 for long UGC-style clips at a lower price per clip

Text, first frame, or Reference

Wan 3.0 appears as two rows in the table. The plain Wan 3.0 row does text-to-video and image-to-video, and image-to-video accepts either a first frame or a first and a last frame. Wan 3.0 Reference takes a set of files, up to 20, and builds a scene from them, which is how you keep the same presenter and the same product across a series of ads. Both rows share the same lengths, resolutions and starting price.

Choose by what you already have. A good product photo and a clear idea of the opening: image-to-video from the first frame. An unboxing or a before-and-after where the ending matters: give it the last frame too, for example a sealed mailer box as the first frame and the linen pillowcases unfolded on a bed as the last, and Wan fills in the motion between. A recurring presenter, several products, a set you want to reuse: Reference. Text-to-video is for ideas you have no images for yet.

Switch off Sound and Prompt enhance when it matters

Wan 3.0 opens with Sound and Prompt enhance both switched on. Sound lets the clip include ambience and speech. Prompt enhance lets the provider rewrite and expand your prompt before the model sees it, which helps a one-line idea and works against you once you are refining specific wording.

We saw both effects in our own runs. In a five-second candle clip made through Blurbly with Wan 3.0 Reference, the brief asked for one continuous shot and no audio. The expanded request split the action into two beats with a cut, the framing changed around the third second, and the file still carried an audio track. For a corrected UGC take we turned Sound and Prompt enhance off, and the clip came back as one continuous scene with no audio stream. Our electronics prompts were all run with rewriting and audio off, so the clips followed the words as written.

A simple rule follows: leave both on while you explore from short prompts, and switch both off when you are iterating on a script or need a silent clip for muted autoplay.

Length, resolution and what you pay

Wan 3.0 is billed per second, at a rate that depends on resolution. The From figure in the table is a 5-second clip at 720p, the default; 480p is cheaper per second, 1080p dearer, and a 30-second take costs roughly six times a 5-second one at the same resolution. Our measured candle clip used the 720p setting.

For a UGC script that suggests drafting at 480p and at the shortest length the spoken line fits into, then rerunning the approved script at 720p or 1080p. Long takes are where Wan 3.0 compares best on price: a 30-second Seedance 2.5 take starts from a higher per-second rate.

Problems we hit, and the fixes

Our first UGC take through Blurbly used a storyboard grid among its references, and the video reproduced the grid: around the third second it turned into a split screen. The correction used a single generated frame as the first-frame input, with no style reference and no storyboard, and came back as one continuous full-frame scene. Keep reference sets to clean, single images of the product, the presenter and the place.

In the same correction the presenter was asked to keep her mouth completely closed, and she still moved her lips and hands, even though the file was silent. If a shot must be both silent and still, frame it on the product rather than on a face, and review every clip for movement you did not ask for before it goes into an ad.