Will the person in the video actually look like my reference?

Yes. Act 2's keyframe always runs character-consistency mode, feeding your reference into the model to align face shape, hairstyle, and build item by item. In the real case the user confirmed "face, hair, and vibe all match" before video rendering started, and the rule is now enforced by the skill.

What are the output specs, and which platforms do they fit?

A 25-second vertical 1080x1920 (9:16) film: a 10s indoor pick-up act plus a 15s outdoor OOTD act, with BGM mixed in. Natively fits TikTok, Instagram Reels, YouTube Shorts, and other vertical feeds.

Why does Act 1 only frame below the calf?

That is the hook of the narrative: a camera about 20 cm off the floor showing only shoes and scattered items, saving the reveal for Act 2's outdoor entrance. A frame with no face also never triggers the video platform's real-person review, so Act 1 renders in one pass.

How is the BGM decided?

Mobbi generates an instrumental BGM to match the outfit's style — a 90s boom bap with heavy drums and vinyl crackle in the real case — and plays it for you first. After you approve, it is mixed in at BGM 0.5 against ambient 0.6, so fabric rustle and footstep SFX still cut through.

Can I rerun just one act if I'm not happy with it?

Yes — the two acts are generated separately and composed at the end. In the real case the user reran only Act 1 as a 10-second single take while Act 2 and the overall structure stayed untouched, then recomposed the final; one video segment takes roughly 8-15 minutes per run.

AI UGC 视频生成器：一张穿搭图直出 25 秒捡衣上身 OOTD Reel

Mobbi 把一张穿搭参考图变成 25 秒两段式 OOTD 种草短片——10 秒低视角捡衣悬念开场，hard cut 切进 15 秒户外多镜头穿搭展示，BGM 混音配齐，无需模特到场、无需摄影团队。

免费试用

使用流程

只需 3 步即可完成。

01 / 03
Upload an Outfit Reference Photo
Provide one outfit photo (optionally plus a person reference). Mobbi decodes the full item list — top, trousers, shoes, bag, and accessories with their colors and materials — then defines the style direction, palette, and a matching outdoor scene. In the real case, one street snap was decoded into seven items and positioned as 90s retro-collegiate x European street.
02 / 03
Lock the Two Opening Keyframes
Act 1's indoor keyframe uses a floor-level composition: camera about 20 cm off the ground, framing only below mid-calf, target items scattered across the floor, shoes strictly matching the reference. Act 2's outdoor keyframe runs character-consistency mode, feeding your reference into the model to lock face, hairstyle, and build into a 1536x2752 vertical frame — video only starts once the likeness is confirmed.
03 / 03
Render on Seedance, Deliver the Final Mix
Seedance 2.0 renders Act 1 as a 10-second single-take pick-up shot and, after clearing the real-face review channel automatically, Act 2 as a 15-second multi-shot OOTD. The two acts are hard-cut together, a MiniMax instrumental BGM is auditioned and then mixed against the ambient SFX, and the 1080x1920, 25-second vertical final is delivered.

你可能还想了解

One outfit reference photo is enough to start: Mobbi decodes every item from it and extracts the person's look. To specify who appears on screen, attach a separate person reference and Act 2 will use it as the character-consistency anchor.