Leo
  • Posts
  • Projects
  • About
  • Posts
  • Projects
  • About
Bilibili抖音小红书

© 2026 XT. All rights reserved.

  1. Home
  2. Projects
  3. Liblib AI | Evoken
  4. Xingliu Migration Guide

Xingliu Migration Guide

An in-product onboarding motion for Xingliu users after platform migration — one continuous flow demonstrating Object Mark tagging, object recognition, and chat references, helping returning users grasp the select → name → reference workflow in the new interface.

Leo WangApril 14, 2026Liblib AI | EvokenAI VideoMotion PackagingProduct Launch
This is a work sample from my previous employment at Liblib AI | Evoken. I served as the video producer and retain the right of attribution under applicable law. Copyright belongs to the original company. This content is for personal portfolio use only, with no commercial or promotional intent.

More projects

  • Lovart: The AI Creative Partner for SMBs

    Lovart: The AI Creative Partner for SMBs

    Brand promo video · AE/PR · Lovart SMB feature launch · storyboard + AIGC (Lovart/Kling) · X/Instagram · 1.11M+ views

  • LibTV 720° Panorama Feature

    LibTV 720° Panorama Feature

    LibTV recently added 720° panoramic video generation — describe a scene in one sentence and AI produces a draggable spherical shot, sharply reducing the setup and iteration cost of immersive spatial storytelling.

  • Lovart Email Section

    Lovart Email Section

    In a marketing email, users glance once — this UI animation uses a fixed prompt bar plus a background asset collage to explain "one prompt, 40 brand assets" in a single frame.

After Xingliu's platform migration, returning users aren't asking "what features disappeared" — they're asking where familiar actions moved and how the interaction syntax changed. This guide motion answers one thing: how Object Mark works, how marks enter the chat, and how two tagged objects become a single executable edit command.

Three Questions the Guide Must Answer

What marking is. Two reference images sit side by side on the canvas — a product shot and a figure shot. The user drops blue mark pins on the shoe and the foot; the system reads each region and surfaces label candidates.

How marks enter chat. Thumbnails and numbers appear at the bottom of the chat input, progressing from generic "image1 / image2" to "1 Shoes" and "2 Feet" — making clear that marks aren't decoration; they're reference variables.

How marks drive editing. Typing "Put this [1 Shoes] on the [2 Feet]" embeds both marks into a natural-language instruction and completes a cross-image object transfer. The full chain from tap to result needs no extra explainer panel.

Production Notes

One operation chain, no segments. Unlike Lovart's five-part onboarding, migration users already have baseline context — they only need the new syntax walked once. Motion follows pin → recognize → name → reference → execute, with the camera locked on canvas–chat linkage throughout.

UI state before copy. Visual feedback lands first when a pin drops; chat updates follow after labels appear. The Object Marked dropdown shows only necessary options to avoid overload. English and Chinese versions share the same AE timeline with text-layer swaps only.

Hold on the instruction. Before the final composite appears, the complete prompt in the input field holds for a beat so viewers can read the mark-to-language mapping, then cut to the result to close.

Overall production & compositing by @Leo Wang