Menu
Use when 用户要把故事镜头做成多宫格/多面板关键帧拼版，再用拼版作为视频参考生成故事驱动视频、广告短片或剧情短片；适合多面板参考图 + 图生视频工作流。
The review-first prompt is selected by default. You can switch to a direct command or download a local copy.
Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
npx skills add https://github.com/LYL1015/JarvisHub --skill flova-multi-panel-videoThe command stays on one line. Scroll horizontally to inspect it before copying.
Prefer a local copy? Download the files currently available to SkillsMP.
Use when the user asks JarvisHub AI Chat or a `webHero` workflow to generate a website, landing page, homepage, brand-style web design, preview-first webpage workflow, or wants section screenshots, webpage assets, and final HTML to be produced as one staged flow.
把 assets/demo 提炼成运行时可用的视觉连续性方法论：先锁资产/角色/场景/镜头语义，再做扩镜与视频。（素材生成场景：角色一致性、视觉锚点锁定、单素材生成前的锚点规范与一次只改一个变量。）
基于用户提供的设计稿图片还原真实网页。该 skill 可用于任意前端项目，但强依赖本地已启动且已授权的 JarvisHub vision 工作流；视觉模型由服务端默认配置决定，支持外部传入 vision prompt，且不可静默降级。
Showing SKILL.md
Based on SOC occupation classification
| name | flova-multi-panel-video |
| description | Use when 用户要把故事镜头做成多宫格/多面板关键帧拼版，再用拼版作为视频参考生成故事驱动视频、广告短片或剧情短片；适合多面板参考图 + 图生视频工作流。 |
用于多宫格关键帧转视频：每个视频镜头先生成一张多格关键帧拼版（multi-panel/contact sheet），再把该拼版作为该镜头视频生成的主要参考，指导内部剪辑与节拍。
This is for story-driven clips where one video shot contains several internal cuts or beats. The multi-panel image is a reference map, not the final video.
imageUrl / videoUrl，不能把提交态当完成态。critic sub-agent：只读取真实媒体并评审，不生成、不补素材。blocked 项，除非本轮工具列表明确暴露对应能力。Use this skill for:
Do not use it when the user only needs a conventional storyboard table, a single keyframe, or a one-shot continuity workflow. Use storyboard-image-production, flova-one-shot-ad-film, or canvas-long-video-production as appropriate.
Follow the JarvisHub Execution Model. Use only tools exposed by this turn's tool list.
Required capability mapping:
If current tools cannot render text labels cleanly inside image cells, use separate metadata in the prompt and output table, but state that visible printed indices may be unreliable.
Pause for confirmation after:
For autonomous runs, replace pauses with critic checks and log decisions.
Prefer a shot that uses the model's reliable clip duration budget and contains multiple internal beats. Do not split a single coherent moment into multiple video shots if it can be represented as one shot with a multi-panel reference sheet.
Each shot should include:
shot_idtarget_aspect_ratioduration_targetscenekey_elementsinternal_cuts: numbered 1..Naudio_layersprompt_notesEach internal cut must include:
Recommended internal cut count: 2-6 per multi-panel image. If a shot has more than 6 important states, merge minor beats or split the shot.
For each video shot, generate one image containing N cells, where N equals the number of internal cuts.
Critical rule:
internal_cuts list.Image requirements:
If visible numbers may pollute final video generation, keep them small, high-contrast, and in a consistent corner, and reinforce in the video prompt that the labels are production metadata.
Use this structure:
Single image contact sheet with [N] separated cinematic frames. Every cell is composed in native [target aspect ratio] to match the final video. Thin neutral dividers between cells. Each cell has a small readable production index [1..N] in the top-left corner matching the storyboard cut list. The spatial layout is flexible and does not define time order. Consistent character/product identity, consistent wardrobe, consistent scene logic, consistent color grade.
Cut 1: [visible moment, framing, action, light, style].
Cut 2: [visible moment, framing, action, light, style].
...
Cut N: [visible moment, framing, action, light, style].
No subtitles, no watermark, no random text, no extra logos.
When references exist, use image-to-image or equivalent this turn's tool list path and name which reference preserves which element.
When the main reference is the multi-panel sheet, start with:
live action
Then describe:
Prompt order:
Template:
live action. Use <<<image_1>>> as the multi-panel reference sheet. Follow the printed cut indices 1 to [N] as the true story order; do not infer time order from the sheet layout. Cut 1 begins with [dynamic description]. Cut 2 continues with [dynamic description]. ... Cut [N] ends with [dynamic description]. Preserve all character/product identities from the reference assets. [camera/style/light rules]. no subtitles, no watermark, no random text, no music.
Avoid precise wall-clock instructions like "at exactly 3.0s." Use ordered cut numbers and relative beat language instead.
If there is a separate narration track, do not put the narration text in the video prompt unless the current model should generate speech.
Bind user-provided assets first:
Generate missing elements only as needed:
Do not generate a separate asset for every cell. The multi-panel sheet already carries per-cut frame states.
Plan audio at storyboard level:
Use wrappers in prompts only if the current video model is expected to produce those sounds:
(...),<...>,{...},【...】.Otherwise, keep video prompts clean and attach/assemble audio separately.
Review the multi-panel image:
Review the generated video:
If the video follows the wrong order, rewrite the prompt to emphasize the cut index order and reduce layout descriptions.
Return:
spec: title, ratio, duration, style, language.storyboard: shot list with internal cut indices.assets: reused/generated element references.multi_panel_sheets: one image URL per shot plus cut count.video_prompts: one prompt per shot.media: generated clip URLs or pending states.audio: planned/generated audio layers.assembly: final video URL or ordered assembly plan.review: sheet and video checks.Do not present a multi-panel sheet as the final video.