Content & Media Creation · skills-shell/superpowers

seedance

Name: seedance
Author: skills-shell/superpowers

Generate videos with ByteDance Seedance 2.0 via inference.sh CLI. Unified model for text-to-video, image-to-video, and reference-to-video with synchronized audio, up to 1080p, 4-15s duration. Pro and Fast variants. Studio variants with private asset library for portrait consistency. Use for: social media videos, music videos, product demos, animated content, AI video with sound. Triggers: seedance, seedance 2, bytedance video, seedance t2v, seedance i2v, seedance r2v, video with audio, seedance 2.0, bytedance seedance, seedance studio

L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandNetwork AccessDual-pass Audit

Author skills-shellGitHub skills-shell/superpowers ↗License MIT（仓库 README 徽标 'License: MIT' 与 .claude-plugin/plugin.json 中 "license": "MIT" 声明；但 pin 树内无 LICENSE 文件，GitHub API 对两 fork 亦报无 license —— 授权只出现在元数据/徽标，无正式许可文本）Commit becc256497

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

A tool skill focused on a single model family: 4 variants (bytedance/seedance-2-0, -fast, -studio, -studio-fast) share a unified invocation surface. Modes are automatically determined by 'which inputs are provided' rather than selected via parameter names, and input exclusivity constraints are explicitly declared (first/last-frame and reference are mutually exclusive).

tools/video/seedance/SKILL.md

The model determines the generation mode from your inputs. These modes are **mutually exclusive** - use either first-frame/last-frame OR reference inputs, not both.

The Modes table clearly details 'automatic model routing': prompt-only=t2v; +image=i2v; +end_image=first-and-last frame; +reference_*=multimodal reference—agents do not need to memorize toggles and can simply assemble input combinations.

The skill embeds a nearly complete API schema: The Parameters table defines types, defaults, and constraints item by item (duration 4-15 or -1 for auto, seven ratio tiers, resolution from 480p to 1080p with Fast capped at 720p, reference_images ≤9 / reference_videos ≤3×15s / reference_audios ≤3×15s with at least one image or video required).

tools/video/seedance/SKILL.md

| `reference_videos` | file[] | - | Reference videos, up to 3. Max 15s each, total max 15s. mp4/mov |

Among the six investigated targets, only this skill provides such an exhaustive parameter contract—it resembles an 'API reference card' more than a plain command snippet collection.

Multimodal prompt references are standardized into formulas: Uses 'type + index' (Image 1/Video 1/Audio 1) to refer to assets within the text while prohibiting asset IDs, and supplies six editing/stitching phrasing templates (add/remove/modify/forward/backward/stitch)—turning 'how to use multimodal references' into reusable patterns.

tools/video/seedance/SKILL.md

Reference assets in your prompt using **type + index**: `Image 1`, `Image 2`, `Video 1`, `Audio 1`. The index is the position within that type in the arrays you provide. Do NOT use asset IDs in prompts.

The 'Video editing formula / extension formula' translates Seedance 2.0's editing and stitching capabilities from raw model features into prompt templates; this is the most 'pedagogical' aspect of this skill compared to other catalog-style skills.

The Studio variants introduce a platform-side asset library: Reference images are automatically uploaded to BytePlus private virtual portrait libraries to achieve cross-video portrait consistency (requiring the safety_identifier user identification parameter)—capabilities and privacy boundaries stem from the same mechanism.

tools/video/seedance/SKILL.md

Studio variants automatically upload reference images to the BytePlus private virtual portrait library for enhanced character consistency

This means portrait assets in Studio mode persist on the BytePlus side to support 'character consistency,' presenting a different asset lifecycle compared to standard mode (a privacy point of note).

Costs are quantified directly inside the skill: The Pricing table provides $4.30-$7.70/M tokens (Fast $3.30-$5.60) and the token formula (width×height×fps×duration)/1024, enabling agents and users to estimate single-video costs prior to invocation.

tools/video/seedance/SKILL.md

Token formula: `(width x height x fps x duration) / 1024`

Similar to the $0.025/s rate in the avatar skill, this repository documents pricing directly inside SKILL.md—cost transparency is a shared selling point across this skill family.

2Core Capabilities

01Text-to-video with synchronized audio (generate_audio defaults to true)

02Image-to-video and start/end frame control (image + end_image boundary frames)

03Multi-image reference maintaining character/clothing/scene (combinatorial addressing with reference_images ≤9)

04In-video element replacement editing (replace/remove/modify phrasing + reference_videos)

05Multi-clip stitching/forward-backward extension (stitch/forward/backward phrasing, ≤3 reference videos)

06Audio reference and vocal timbre guidance (reference_audios, requires at least one image/video)

07Studio portrait consistency (private asset library + safety_identifier)

3External Dependencies

Kind	Dependency / Endpoint
cli	belt（inference.sh CLI）
package	belt-sh/cli（npx skills add 安装）
api	ByteDance Seedance 2.0 平台 app（bytedance/seedance-2-0、-fast、-studio、-studio-fast；Studio 模式联动 BytePlus 私有肖像库）
network	平台文档（docs/apps/running、streaming）
network	安装说明托管
package	相关技能包 inference-sh/skills@ai-video-generation / google-veo / ai-image-generation / ai-avatar-video

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Graded/per-use authorization, sandbox recommended

For Administrators

Approval workflow, rollback support, full audit

Evidence & Triggers (File Paths & Excerpts)

网络端点: Declared external network access: platform belt CLI → inference.sh platform → ByteDance Seedance/BytePlus (generation and Studio
SKILL.md: Declared invocation of host CLI interpreter or external runtime environment

Audit Notes: Pure documentation-style + whitelisted belt execution: Contains no scripts, credential-reading routines, or direct-connection networking code; outbound targets are predictable (inference.sh → ByteDance). Points of caution: 1) Studio variants automatically upload reference images to BytePlus private portrait libraries, requiring prior evaluation of third-party retention for portrait and copyrighted assets; 2) Billed by tokens ($4.30-$7.70/M tokens), making high-resolution, long-duration runs noticeably expensive per generation; 3) Referenced assets (videos/audios ≤15s×3) leave the network entirely; 4) Supply chain factors match previous entries (unpinned npx, unrenamed inference-sh mirror, no in-tree LICENSE). Per unified tiering guidelines: The `belt` invoked by this skill is a CLI distributed via third-party npm/npx (installed via `npx skills add belt-sh/cli`), runs under a logged-in `belt login` state, and consumes inference.sh account quota/billing—falling under 'invokes third-party CLI and touches quota/login state', and is therefore rated Orange (use after evaluation); the skill source code itself contains no token or secret reading logic.

5Second-Pass Verification

[ok] description 数字声明（1080p、4-15s、音频同步）与正文一致 — Parameters 表 'Duration in seconds (4-15), or -1 for auto'、'resolution ... 1080p (Fast: 480p, 720p only)'、generate_audio 默认 true 均与 description 相符。
[ok] Pro/Fast/Studio 四变体模型表 — Models 表 4 行（2.0 / 2.0 Fast / 2.0 Studio / 2.0 Studio Fast）与 description 一致；Studio 自动上传 BytePlus 的行为有正文专段表述。
[ok] Modes 互斥约束与示例一致 — 各示例要么用 image/end_image 组合要么用 reference_* 组合，无混用；正文明确 'mutually exclusive'。
[ok] 引用上限 9/3/3 与示例用量 — Parameters 表 'reference_images, up to 9'、'reference_videos, up to 3'、'reference_audios, up to 3'；示例最大用量 3 图+1 视频+1 音频、3 视频，均未超限。
[ok] prompt 引用公式（type+index、禁 asset ID） — Prompt Guide 原文 'Do NOT use asset IDs in prompts' 与全部多模态示例的 Image N/Video N/Audio N 写法一致。
[ok] safety_identifier 参数 — Parameters 表 'Unique end-user identifier for safety policy (max 64 chars, hash of user ID recommended)' 与 Studio 示例中 'user-abc123' 对应。
[ok] 无脚本/无凭证/无直连网络 — 目录仅 SKILL.md；curl/wget/密钥 token 零命中；无 shell 重定向；执行面 allowed-tools: Bash(belt *)。
[ok] license/stars/commit 元数据 — pin == 本地 HEAD == becc25649700；skills-shell/superpowers stars=0、pushed_at 2026-09-02T14:16:56Z；MIT 声明见 plugin.json/README 徽标，树内无 LICENSE 文件。

6Verdict & Scenarios

Most comprehensive single-model reference: Fully covers t2v, i2v, first/last frame, multimodal referencing, editing, stitching, extension, and audio referencing within a single skill, backed by strict input contracts (Modes exclusivity + Parameters table) to minimize invocation errors.

Pedagogical prompt formulas make model techniques explicit: Image N/Video N/Audio N addressing combined with add/remove/modify/stitch phrasing spares agents from trial-and-error.

Pre-computable costs: The token formula and unit price table allow quick cost estimations prior to invocation.

Studio variants offer an official path for 'cross-video portrait consistency' via a private asset library, providing more stability than repeatedly referencing images in prompts.

Best for: Ideal for users committed to producing videos with Seedance 2.0: content teams requiring character consistency (Studio), audio-video synchronization, and advanced editing/stitching capabilities; also suitable for agent authors seeking a 'complete parameter reference manual for this model' beyond the generic ai-video-generation entry point.

Not for: Not suitable for exploratory users needing horizontal comparisons across multiple models (who should use the general ai-video-generation entry point first); not suitable for scenarios where assets cannot leave the local network or where BytePlus asset library retention is a concern; not suitable for zero-budget batch automation; Studio portrait consistency scenarios require prior verification of portrait licensing and third-party data retention policies.

Install Agent-Ready

① Verified Mirror 2026-09-09

Option A · Direct DownloadDownload seedance.tar.gz

sha256: ebc2e07ba23fe046…

Option B · Copy Guide for Agent

install.json

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit becc256497.