atlas-cloud
calesthio/OpenMontage · Agent Skill
透過 Atlas Cloud 閘道生成和編輯影象和影片,支援各種 AI 媒體模型。
前往安裝 ↓總覽
| 發布者 | |
|---|---|
| 儲存庫 | calesthio/OpenMontage ↗ |
| 技能資料夾 | atlas-cloud |
| 星 | 58,214 |
| 分支 | 7,332 |
| 附帶檔案 | 1 |
| 授權 | AGPL-3.0 |
| 最近更新 | 2026年9月6日 |
| 連結 | 技能原始碼 ↗ · 儲存庫 ↗ |
Mozgi 檢查
我們在技能檔案中發現的內容。詳細資訊需訂閱。
安裝
技能就是一個含 SKILL.md 的資料夾。選擇你的代理,在專案根目錄的終端機(macOS、Linux 或 Windows 的 Git Bash)執行指令。
把提示貼到 Claude、Codex 或其他助理:它會自行檢查並安裝技能。
Install the “atlas-cloud” skill from mozgi.io (https://mozgi.io/skills/atlas-cloud/) for me. 1. Download the archive: https://mozgi.io/skills/atlas-cloud/download If it fails, take the skill folder from the source: https://github.com/calesthio/OpenMontage/tree/main/.agents/skills/atlas-cloud 2. Read SKILL.md and the other files. Check the skill is safe: no hidden commands, no sending data elsewhere, no scripts doing something other than described. If anything looks suspicious, stop and tell me. 3. Put the “atlas-cloud” folder into your agent's skills folder: .claude/skills/ for Claude Code, .agents/skills/ for Codex, .cursor/skills/ for Cursor, .gemini/skills/ for Gemini CLI. Inside the project if we are in one, otherwise in the home folder (~/.claude/skills/ etc.). 4. Make sure <skills folder>/atlas-cloud/SKILL.md exists. 5. Briefly tell me what this skill does and offer to try it on my task. If you cannot do a step yourself, tell me what to click or type.
git clone --depth 1 --branch main https://github.com/calesthio/OpenMontage.git /tmp/OpenMontage mkdir -p .claude/skills cp -r /tmp/OpenMontage/.agents/skills/atlas-cloud .claude/skills/atlas-cloud
重新啟動 Claude Code — 技能就會出現在清單中,手動呼叫:/atlas-cloud。要套用到所有專案,請放在 ~/.claude/skills。
git clone --depth 1 --branch main https://github.com/calesthio/OpenMontage.git /tmp/OpenMontage mkdir -p .agents/skills cp -r /tmp/OpenMontage/.agents/skills/atlas-cloud .agents/skills/atlas-cloud
Codex 會從目前資料夾往上到儲存庫根目錄,在 .agents/skills 中尋找技能;用 /skills 或 $atlas-cloud 呼叫。所有專案:~/.agents/skills。
git clone --depth 1 --branch main https://github.com/calesthio/OpenMontage.git /tmp/OpenMontage mkdir -p .cursor/skills cp -r /tmp/OpenMontage/.agents/skills/atlas-cloud .cursor/skills/atlas-cloud
Cursor 啟動時會載入技能;在 Agent 聊天輸入「/」呼叫。Cursor 也會讀取 .claude/skills 與 .agents/skills。所有專案:~/.cursor/skills。
git clone --depth 1 --branch main https://github.com/calesthio/OpenMontage.git /tmp/OpenMontage mkdir -p .gemini/skills cp -r /tmp/OpenMontage/.agents/skills/atlas-cloud .gemini/skills/atlas-cloud
任務符合時,Gemini CLI 會提議啟用技能,由你確認。所有專案:~/.gemini/skills。
安裝前請先檢視 SKILL.md 與相關檔案。點擊檔案即可預覽。
SKILL.md 內容
作者原文 · 唯讀Atlas Cloud
Route complete productions through image_selector or video_selector. Call
atlas_image or atlas_video directly when the user names Atlas Cloud or an
exact Atlas model. Never substitute a direct vendor endpoint for an Atlas request.
Set ATLASCLOUD_API_KEY. The aliases ATLAS_CLOUD_API_KEY and ATLAS_API_KEY
are accepted for compatibility.
Preflight every paid call
1. Read tool.get_info()["model_catalog"]; do not infer availability from a
search collection or invent a task suffix.
2. Select an exact model id and operation supported by that catalog.
3. Announce the Atlas tool, provider, exact model, request count, and estimated
cost before submitting.
4. Use local-path inputs when convenient. The tool uploads them through Atlas.
5. Save outputs inside the active projects/<project-id>/ tree and inspect them.
Video routes
| Family | Operations | Exact route notes |
|---|---|---|
| Seedance 2.5 | text, image, reference | bytedance/seedance-2.5/{text,image,reference}-to-video; image mode uses image plus optional last_image; reference mode accepts up to 30 images, 10 videos, and 10 audios (including audio-only); 4–30s; $0.134/s |
| Seedance 2.0 | text, image, reference | bytedance/seedance-2.0/{text,image,reference}-to-video; 4–15s; $0.112/s |
| Gemini Omni Flash | text, image, reference, video edit | Standard routes use google/gemini-omni-flash/...; developer routes exist for text/image/reference only. Standard image mode uses one image; standard reference uses images; developer reference requires one video_clips object. |
| MiniMax H3 | text, image, reference | minimax/h3/{text,image,reference}-to-video; image mode supports optional end_image; reference mode requires refers objects; 4–15s; 768P/2K; $0.10/s |
Use canonical OpenMontage fields:
tool.execute({
"prompt": "...",
"model": "minimax/h3/reference-to-video",
"operation": "reference_to_video",
"duration": 10,
"resolution": "2K",
"reference_images": ["projects/demo/character.png"],
"reference_audios": ["projects/demo/performance.wav"],
"output_path": "projects/demo/h3.mp4",
})
For Gemini developer reference video, pass video_clips objects with url,
start, and ends. For H3, callers may pass normalized reference_images,
reference_videos, and reference_audios; the tool converts them to refers.
Do not silently clamp duration, resolution, or ratio. Invalid values must fail
before billing with the supported choices in the error.
Image routes
| Family | Operations | Exact route notes |
|---|---|---|
| Seedream 5.0 Pro | generate, edit, decompose | bytedance/seedream-v5.0-pro/text-to-image, /edit, /layer-decomposition; edit accepts up to 10 images; sizes use WIDTH*HEIGHT |
| Seedream 5.0 Lite | edit | bytedance/seedream-v5.0-lite/edit; no live Lite text-to-image sibling is assumed; accepts up to 14 images |
| GPT Image 2 | generate, edit | openai/gpt-image-2/text-to-image, /edit; sizes use WIDTHxHEIGHT; edit accepts up to 10 images |
| Nano Banana 2 | generate, edit | google/nano-banana-2/text-to-image, /edit; uses aspect ratio plus 1k/2k/4k resolution; edit accepts up to 14 images |
Set generation_mode to generate, edit, or decompose. Source images can
be supplied as image_path, image_paths, image_url, or image_urls.
tool.execute({
"prompt": "Replace the packaging with matte cobalt glass; preserve the logo",
"model": "google/nano-banana-2/edit",
"generation_mode": "edit",
"image_paths": ["projects/campaign/source.png"],
"resolution": "2k",
"output_path": "projects/campaign/revised.png",
})
Result and failure contract
Both tools submit to Atlas, poll the returned prediction id, download every
output, and return ToolResult provenance containing the provider, exact model,
prediction id, submitted parameters, source URL, artifact paths, and estimated
cost. A missing key, unsupported route, invalid enum, missing media input, failed
prediction, timeout, or download failure must return a failed result without
switching providers.
Confirm current pricing and schemas from the machine-readable model page
(Accept: text/markdown) immediately before quoting or spending on a batch.
常見問題
atlas-cloud 技能有什麼用途?
透過 Atlas Cloud 閘道生成和編輯影象和影片,支援各種 AI 媒體模型。
如何安裝 atlas-cloud?
打開本頁的「安裝」區塊並選擇你的代理 — Claude Code、Codex、Cursor 或 Gemini CLI:那裡有把技能資料夾複製到正確位置的現成指令。也可以複製「透過助理」分頁中的提示,讓助理自行檢查並安裝技能。
atlas-cloud 適用哪些 AI 模型?
技能是純 Markdown 指示(Agent Skills 格式),因此 atlas-cloud 適用任何讀取 SKILL.md 的代理:Claude Code、Codex、Cursor、Gemini CLI、GitHub Copilot、OpenCode 等。只有在任務符合技能描述時,模型才會載入指示。
atlas-cloud 是免費的嗎?
此技能由 calesthio 公開發布,採用 AGPL-3.0 授權。技能本身免費 — 你只需向 AI 供應商支付實際使用的 token 費用。
安裝技能安全嗎?
技能是代理會遵循的指示,安裝前請閱讀 SKILL.md 及相關檔案。技能資料夾中的腳本只有在代理執行時才會運行 — 留意它的動作,不要授予多餘權限。
技能和 MCP 伺服器或外掛有何不同?
MCP 伺服器或外掛為模型提供新工具 — 會執行並回傳結果的程式碼。技能提供的是知識與流程:如何處理任務、要遵循哪些步驟、好的成果長什麼樣子。它不需要伺服器或 API 金鑰。