Agent Skills › gooseworks-ai/goose-skills › create-video-fal

create-video-fal

GitHub

通过 GooseWorks 代理调用 FAL 视频模型(如 Kling、Veo),实现图文/文生视频生成,支持参数配置、结果下载及合规审查,用于制作视频广告基础素材。

skills/video/create-video-fal/SKILL.md gooseworks-ai/goose-skills

触发场景

需要生成图像转视频或文本转视频内容 使用特定 FAL 模型创建视频广告片段 需要处理视频生成的合规性与角色一致性检查

安装

npx skills add gooseworks-ai/goose-skills --skill create-video-fal -g -y
更多选项

非标准路径

npx skills add https://github.com/gooseworks-ai/goose-skills/tree/main/skills/video/create-video-fal -g -y

不安装直接使用

npx skills use gooseworks-ai/goose-skills@create-video-fal

指定 Agent (Claude Code)

npx skills add gooseworks-ai/goose-skills --skill create-video-fal -a claude-code -g -y

安装 repo 全部 skill

npx skills add gooseworks-ai/goose-skills --all -g -y

预览 repo 内 skill

npx skills add gooseworks-ai/goose-skills --list

SKILL.md

Frontmatter
{
    "name": "create-video-fal",
    "status": "active",
    "description": "Image-to-video (or text-to-video) via any FAL video model (Kling, Seedance, Veo), routed through the GooseWorks proxy, so the call is charged to your GooseWorks credits. The template recipe names the model + params; image_url inputs must be public URLs (the orchestrator hosts local frames with the MCP media_upload). Returns the result video URL and downloads it. Use for the generative base clip of any video-ad format."
}

create-video-fal

Image-to-video (or text-to-video) via any FAL video model (Kling, Seedance, Veo), routed through the GooseWorks proxy, so the call is charged to your GooseWorks credits. The template recipe names the model + params; image_url inputs must be public URLs (the orchestrator hosts local frames with the MCP media_upload). Returns the result video URL and downloads it. Use for the generative base clip of any video-ad format.

Run

gen_video.py --model fal-ai/kling-video/v3/pro/image-to-video --payload '{...}' --out clip.mp4 — charged to your GooseWorks credits; host-swaps the FAL queue URLs; downloads the result.

Contract

  • Paid calls route through the GooseWorks proxies (charged to your GooseWorks credits) via the bundled media_proxy.py — never a provider SDK's default host.
  • The template recipe (DB) supplies the model + params; this capability is generic.

Rejection, physical constraints and cast planning

A provider likeness/policy rejection stops the attempt. Preserve the provider's reason, request id and charged/uncharged/unknown state. Do not resubmit an identical rejected payload. Offer a permitted original character, user-cleared reference, or a supported non-likeness route only when allowed by that provider. A different model is not a policy bypass. Review changed inputs and extra spend through the normal approval flow.

Before generation, write a scene checklist from the brief: each wearable's exact count and body location; which hand holds each object; allowed gestures; object contacts and movement; cast identities and reference ownership. Keep unnecessary hands still, use one simple action per shot, and review the whole generated take against the checklist. A prompt is prevention, not proof: reject extra/missing products, impossible contacts or identity drift.

For multiple characters, compare a shared scene with pinned references, fewer people per shot, and separately generated/composed plates. The first preserves interaction but risks identity drift; separate plates improve control but add composition work and may weaken interaction. Lock an approved reference per person and map who speaks each line. No six-character/two-attempt guarantee is supported. A future paid benchmark must state cast size, attempts, budget, model/settings and pass criteria (identity, speaker, counts, gestures and complete dialogue) and retain every failure.

Model notes

How the video models behave, measured on shipped projects. Each note lives here once; recipes point here instead of repeating it. Seedance and the talking-creator models keep their notes in their own atoms.

  • Veo 3.1 reads states as stills. "Legs in jeans on a wet curb" gives a near-static clip. Name actions with verbs (steps off, pivots, taps) when the clip has to move.
  • Veo 3.1 keeps the start image's composition (angle, distance, what is cropped) for the whole clip. The start image is the framing, not just a reference.
  • Veo 3.1 takes 4, 6 or 8 seconds only. It paces speech to fill the length, so ask for about 6 rather than 8 and keep the tail short; defects live in the tail. Handheld motion and a music bed are luck per seed, so plan several seeds for a shot that needs them.
  • Veo 3.1 Fast ad-libs words and, from about 4 seconds, the eyes can widen and stare. Ask for 4 seconds for a one-line clip.
  • Kling v3 for flat 2D or editorial illustration: at cfg_scale 0.5 or lower, with motion-only prompts, it adds on-style motion where Seedance and Veo invent photoreal middle states.
  • Kling v3 holds one state per clip. Staged "first, then, finally" prompts change only in the last half-second, and a neutral face drifts to a smile and then to mouthing words. Ask for one state per clip and build the arc in the edit.
  • Kling v3 standard returns 2:3 (784x1176) when asked for 9:16. Check the returned size and give the start image side margin when Kling is the engine.
  • Hard constraints need saying three ways (image and video). A single "no face" holds about half the time. State it as a positive rule near the top, a negation in the middle and a scope rule at the end.
  • Failures: an NSFW false positive needs the visual trigger words removed; a timeout with no detail gets one unchanged retry, then a simpler prompt; a rate limit means wait. A policy refusal is final (above).
  • Realism comes from real pixels. Prompted "grainy, shot on a phone" still reads as generated. When a shot must look filmed, restyle real footage, and narrow what the model invents: a blank glowing screen generates well, a legible interface does not, so composite the real UI in.
  • Change over a long period cannot be generated. A shot defined by change longer than one clip (screens changing through a work session) needs real footage; the model renders one moment.

版本历史

  • 1add09b 当前 2026-11-02 21:55

同 Skill 集合

docs/skill-template/SKILL.md
skills/entry/goose-ads/SKILL.md
skills/entry/goose-product-photos/SKILL.md
skills/entry/goose-video/SKILL.md
skills/entry/gooseworks/SKILL.md
skills/entry/make-custom-video/SKILL.md
skills/marketing/analyze-performance-call-and-learn/SKILL.md
skills/marketing/analyze-performance-diagnose/SKILL.md
skills/marketing/analyze-performance-intake/SKILL.md
skills/marketing/analyze-performance-rank-and-read/SKILL.md
skills/marketing/analyze-performance-revenue-join/SKILL.md
skills/marketing/claims-gate/SKILL.md
skills/marketing/email-lifecycle-broadcast-planner/SKILL.md
skills/marketing/email-lifecycle-flow-writer/SKILL.md
skills/marketing/email-lifecycle-lifecycle-audit/SKILL.md
skills/marketing/email-lifecycle-offer-math/SKILL.md
skills/marketing/email-lifecycle-send-check/SKILL.md
skills/marketing/honest-review/SKILL.md
skills/marketing/improve-search-visibility-citation-check/SKILL.md
skills/marketing/improve-search-visibility-entity-footprint/SKILL.md
skills/marketing/improve-search-visibility-fix-writer/SKILL.md
skills/marketing/improve-search-visibility-page-and-foundation-check/SKILL.md
skills/marketing/improve-search-visibility-question-set/SKILL.md
skills/marketing/improve-search-visibility-store-listing-audit/SKILL.md
skills/marketing/make-content-content-brief/SKILL.md
skills/marketing/make-content-source-to-units/SKILL.md
skills/marketing/make-content-two-pass-edit/SKILL.md
skills/marketing/make-static-ads-ad-copy/SKILL.md
skills/marketing/make-static-ads-claims-extract/SKILL.md
skills/marketing/make-static-ads-compose/SKILL.md
skills/marketing/make-static-ads-concept-grid/SKILL.md
skills/marketing/make-static-ads-copy-variants/SKILL.md
skills/marketing/make-static-ads-find-references/SKILL.md
skills/marketing/make-static-ads-fix-rejected/SKILL.md
skills/marketing/make-static-ads-graphic-request/SKILL.md
skills/marketing/make-static-ads-render-review/SKILL.md
skills/marketing/marketing-guide/SKILL.md
skills/marketing/outcomes-ledger/SKILL.md
skills/marketing/plan-campaign-brief-and-assets/SKILL.md
skills/marketing/plan-campaign-budget-and-tests/SKILL.md
skills/marketing/plan-campaign-channel-decision/SKILL.md
skills/marketing/plan-campaign-goal-chain/SKILL.md
skills/marketing/research-audience-account-sourcing/SKILL.md
skills/marketing/research-audience-fit-definition/SKILL.md
skills/marketing/research-audience-people-and-contacts/SKILL.md
skills/marketing/research-audience-score-segment-qa/SKILL.md
skills/marketing/research-audience-segment-pass/SKILL.md
skills/marketing/research-audience-signals-and-hooks/SKILL.md
skills/marketing/research-competitors-category-pass/SKILL.md
skills/marketing/research-competitors-edge-cases/SKILL.md

元信息

文件数
0
版本
1add09b
Hash
63000e9c
收录时间
2026-11-02 21:55

首页 - Wiki
Copyright © 2011-2026 iteam. Current version is 2.156.0. UTC+08:00, 2026-11-18 19:55
浙ICP备14020137号-1