ui-showreel-forge

UI Showreel Forge

A complete production workflow from UI design mockups to commercial-grade motion video.

Core methodology:Chained reference-image method. By using previously generated images as physical reference inputs for subsequent frames, it achieves cross-frame style consistency in AI image generation. It is paired with "establishing frame / effect frame" storyboard planning and a director script format that can be used directly with image-to-video models such as Kling and Omni.

When to use

Use this skill in the following scenarios:

• You already have a set of UI design mockups and want to turn them into a high-quality showreel, product promo, or motion showcase video
• You need to first produce style-consistent keyframes/reference images, then hand them to video models such as Kling and Omni to generate motion clips
• You want to upgrade "turning screenshots into video" from guesswork-style prompting into a stable, reusable production process

Default toolchain

This skill is currently designed by default around the following combination:

• Stitch MCP: Generate or batch-acquire UI screenshots
• 图像生成模型: Generate establishing frames/effect frames based on style reference images and the chained reference-image method
• 可灵 / Kling or Omni: Consume keyframes and the director script to generate the final motion video

If you do not use Kling, you can still reuse this method as long as the target model supports multi-image input, image-to-video, or director-script-style control.

Typical inputs

• UI screenshots generated by Stitch MCP
• 1 style reference image
• Optional negative example images
• Target video aspect ratio, pacing, and shot style requirements

Typical outputs

• N style-consistent showreel reference-image sequences
• An establishing frame/effect frame storyboard table
• A director script that can be pasted directly into models such as Kling and Omni

Overall process

[Phase 0] Stitch MCP 生成 UI 设计稿
    ↓
[Phase 1] 素材准备：截图 + 风格锚定参考图
    ↓
[Phase 2] 分镜规划：建立帧/效果帧交替 + 过渡逻辑
    ↓
[Phase 3] 链式图像生成：锚点帧 → 建立帧 → 效果帧（DAG 顺序执行）
    ↓ 每帧生成后立即目视自检，不合格立即返工
[Phase 4] 自检与返工
    ↓
[Phase 5] 导演脚本编译（面向可灵/Kling、Omni 等模型）

Hard constraints

• Have a complete storyboard first, then start generating images
• Each effect frame must be preceded by a corresponding establishing frame
• All effect frames must continue to be generated based on the corresponding establishing frame, rather than going back to the original screenshot for reconstruction
• Visually inspect immediately after each generation; do not accumulate problems until the end

Reference documents

• references/phase1_material_prep.md: Material preparation and style anchoring
• references/phase2_storyboard.md: Storyboard planning and visual blind-spot design
• references/phase3_chain_reference.md: Chained reference-image generation methodology
• references/phase4_self_check.md: Self-check and rework protocol
• references/phase5_omni_script.md: Director script compilation template