功能描述

a2e.ai full API: Image Gen (Text2Image, NanoBanana, GPT Image, Flux 2), Video Gen (Image2Video with LoRA/FLF2V support, Video2Video, Kling 3.0, Wan 2.6, Sora...

使用说明 (SKILL.md)

a2e.ai — Full Platform Skill

Name: a2e.ai Full Platform
Author: danielgrobelny

Complete API access to a2e.ai. A2E-eigene Modelle sind kostenlos für max-user. Premium-Modelle (Wan 2.6, Kling 3.0, Seedance 1.5 Pro etc. — erkennbar am 🔥) kosten Coins.

Auth

source ~/.openclaw/workspace/.env  # loads A2E_KEY

If key expired → Daniel generates new one at https://video.a2e.ai → Account Settings → API Keys

Base URL: https://video.a2e.ai Auth header: Authorization: Bearer $A2E_KEY

Quick CLI — a2e.sh

{baseDir}/scripts/a2e.sh balance                            # Check coins
{baseDir}/scripts/a2e.sh generate "prompt" [WxH] [style]   # Text2Image (general|manga)
{baseDir}/scripts/a2e.sh nano "prompt" [image_url]          # NanoBanana/Gemini
{baseDir}/scripts/a2e.sh faceswap \x3Cface_url> \x3Ctarget_url>   # Face Swap
{baseDir}/scripts/a2e.sh headswap \x3Chead_url> \x3Ctarget_url>   # Head Swap
{baseDir}/scripts/a2e.sh img2vid \x3Cimage_url> "prompt"       # Image to Video
{baseDir}/scripts/a2e.sh vid2vid \x3Cimage_url> \x3Cvideo_url> "prompt"  # Video to Video
{baseDir}/scripts/a2e.sh tts "text" [voice_id]              # Text to Speech
{baseDir}/scripts/a2e.sh voices                             # List TTS voices
{baseDir}/scripts/a2e.sh voiceclone "name" \x3Caudio_url>      # Clone a voice
{baseDir}/scripts/a2e.sh avatar \x3Canchor_id> \x3Caudio_url>     # AI Avatar video
{baseDir}/scripts/a2e.sh avatars                            # List available avatars
{baseDir}/scripts/a2e.sh talkphoto \x3Cimage_url> \x3Caudio_url>  # Talking Photo
{baseDir}/scripts/a2e.sh talkvideo \x3Cvideo_url> \x3Caudio_url>  # Talking Video
{baseDir}/scripts/a2e.sh dub \x3Cvideo_url> \x3Ctarget_lang>      # AI Dubbing
{baseDir}/scripts/a2e.sh caption \x3Cvideo_url>                # Remove captions
{baseDir}/scripts/a2e.sh tryon \x3Cperson> \x3Cmask> \x3Ccloth> \x3Ccloth_mask>  # Virtual Try-On
{baseDir}/scripts/a2e.sh createavatar "name" \x3Curl> [type]   # Create custom avatar (video|image)
{baseDir}/scripts/a2e.sh trainlipsync \x3Cavatar_id>           # Train Studio lip-sync on avatar
{baseDir}/scripts/a2e.sh cloneavatarvoice \x3Cavatar_id>       # Clone voice from avatar video
{baseDir}/scripts/a2e.sh myavatars                          # List custom avatars
{baseDir}/scripts/a2e.sh removeavatar \x3Cavatar_id>           # Delete custom avatar
{baseDir}/scripts/a2e.sh addface \x3Cface_image_url>           # Save face for reuse
{baseDir}/scripts/a2e.sh myfaces                            # List saved faces
{baseDir}/scripts/a2e.sh facepreview \x3Cface> \x3Ctarget>        # Quick face swap preview
{baseDir}/scripts/a2e.sh backgrounds                        # List avatar backgrounds
{baseDir}/scripts/a2e.sh addbg \x3Cimage_url>                  # Add custom background
{baseDir}/scripts/a2e.sh upload \x3Curl>                       # Save URL to a2e storage
{baseDir}/scripts/a2e.sh presign                            # Get presigned upload URL
{baseDir}/scripts/a2e.sh languages                          # List available languages
{baseDir}/scripts/a2e.sh status \x3Cid> \x3Cengine>               # Check task status
{baseDir}/scripts/a2e.sh poll \x3Cid> \x3Cengine>                 # Poll until done

Available Engines (engine names for status/poll)

Engine	CLI name	Endpoint prefix
Text2Image	t2i	userText2image
NanoBanana	nano	userNanoBanana
Face Swap	faceswap	userFaceSwapTask
Head Swap	headswap	(TBD)
Image to Video	img2vid	userImage2Video
Video to Video	vid2vid	motionTransfer
Avatar Video	avatar	video
Talking Photo	talkphoto	talkingPhoto
Talking Video	talkvideo	talkingVideo
AI Dubbing	dub	userDubbing
Caption Removal	caption	userCaptionRemoval
Virtual Try-On	tryon	virtualTryOn
Voice Clone	voiceclone	userVoice

Common Async Pattern

All tasks follow the same flow:

POST .../start → get _id + current_status: "initialized"
Poll via GET .../{_id} or .../allRecords until completed
Status: initialized → processing → completed | failed
Result URLs (image_urls, result_url, video_url) expire after ~3 days
On failure: check failed_message + failed_code

Known Quirks

NanoBanana GET by _id returns 404 → use allRecords to find tasks
Text2Image input_images does NOT work → use NanoBanana for reference images
German text in Text2Image often has typos → NanoBanana (Gemini) handles it correctly
Result URLs are signed and expire after ~3 days → download/save promptly

Additional Models (Web UI, API endpoints unconfirmed)

These are available on the platform but may not have documented API endpoints yet:

Kling 3.0 — Cinema-grade video (text/image/motion modes, sound gen)
Wan 2.6 — Cinematic video with synced audio
Sora 2 Pro — OpenAI video model
Seedance 1.5 Pro — Multi-shot video
Veo 3.1 — Google video generation
GPT Image 1.5 — OpenAI image model
Flux 2 Pro — Black Forest Labs, speed-optimized
Grok Imagine — xAI image/video
Nano Banana 2 — Next-gen NanoBanana
Photobook — Multiple portraits from one photo (relevant for AI Photoshooting DAN-102)

Full API Reference

See {baseDir}/references/api-complete.md for all endpoints with request/response details.

安全使用建议

This skill is coherent with its stated purpose (it wraps the a2e.ai API) and only needs an A2E API key — but take these precautions before installing/using it: - Review ~/.openclaw/workspace/.env: the SKILL.md tells the agent to source that file. If that file contains any other secrets (AWS keys, tokens, etc.), sourcing it will load them into the agent environment and could unintentionally expose them. Prefer to store only A2E_KEY in the file or change the instructions to load A2E_KEY explicitly. - Verify local tool availability: the included script uses curl and jq. Ensure those are present and from trusted sources; the skill metadata does not list them as required binaries. - Understand privacy/legal implications: features like face-swap, voice-clone, and avatar generation upload images/audio to a remote service. Make sure you have rights/consent for any media you process. - Limit the A2E_KEY scope and rotate it if possible: give the skill an API key with minimal permissions and short lifetime if the platform supports it, and rotate the key after initial testing. - Monitor network activity and API usage (coins/credits): because the agent can call the API, unexpected autonomous calls could consume credits if the key is compromised or the skill is misused. If you want the skill but worry about the .env sourcing, ask the publisher (or modify the script locally) so the agent only reads the single A2E_KEY value instead of sourcing the whole file.

功能分析

Type: OpenClaw Skill Name: a2e-image Version: 2.1.0 The a2e-image skill bundle provides a comprehensive interface for the a2e.ai media generation platform, including image/video generation, face swapping, and TTS. The primary logic is contained in scripts/a2e.sh, which uses curl and jq to interact with the official https://video.a2e.ai API in a secure manner, utilizing environment variables for authentication as per OpenClaw standards. No evidence of data exfiltration, malicious command execution, or harmful prompt injection was found; the bundle is well-documented and its behavior aligns strictly with its stated purpose.

能力评估

ℹ Purpose & Capability

The name, description, SKILL.md, reference docs, and script all align with an a2e.ai image/video/voice API helper; requesting A2E_KEY is appropriate. Minor mismatch: the provided CLI script uses common binaries (curl, jq, date) but the skill's metadata declares no required binaries. This is an incoherence (not necessarily malicious) that could cause runtime failures.

⚠ Instruction Scope

SKILL.md explicitly instructs the agent to run `source ~/.openclaw/workspace/.env` to load A2E_KEY. Sourcing the entire .env file can expose any other environment variables or secrets present there (not just the declared A2E_KEY). The script performs many network calls (POST/GET to video.a2e.ai endpoints) which is expected for this API, but the explicit sourcing of a full env file is a scope expansion and risk for secret exposure.

ℹ Install Mechanism

There is no install spec (instruction-only), so nothing is downloaded or written during install — good. The included scripts will be executed if used. The script relies on jq and curl; the skill metadata does not declare these required binaries, creating a functional mismatch but not a direct supply-chain red flag.

⚠ Credentials

The only declared credential is A2E_KEY, which is proportional to the stated functionality. However, the runtime instruction to source ~/.openclaw/workspace/.env may load additional environment variables without declaring them, potentially exposing unrelated secrets. No other unrelated credentials are listed in requires.env.

✓ Persistence & Privilege

The skill does not request always:true and uses default invocation permissions. It does not attempt to modify other skills or system-wide config. Autonomous invocation is allowed by default (disable-model-invocation=false) — this is normal for skills, and here it is not combined with additional dangerous privileges.

版本历史

v2.1.0

Add complete pricing reference (scraped from live_pricing page). All features with credit costs documented.

v2.0.0

Major: Verified all 70 endpoints against official OpenAPI spec from Apidog MCP server. Fixed: createavatar (startTraining + gender), trainlipsync (continueTranining), cloneavatarvoice (addVoiceClone), myavatars (records), myfaces (records), voiceclone (training + model/language/gender params, 4 engines: a2e/cartesia/minimax/elevenlabs), TTS (cloned voice support with country/region), backgrounds (custom_back), presign (r2). Added: watermark, avatarstatus, avatarvideos, detectlang, facepreviewstatus. Raw OpenAPI spec saved in references/.

v1.1.1

Remove personal info from skill description

v1.1.0

Fix: A2E models free for max-users (not all models). Add img2vid params: model_type (GENERAL/FLF2V), end_image_url, lora support. Correct pricing info.

v1.0.0

Initial release: Complete a2e.ai API coverage — Image Gen (Text2Image, NanoBanana), Video (img2vid, vid2vid), Face/Head Swap, TTS + Voice Clone, AI Avatars, Talking Photo/Video, AI Dubbing, Caption Removal, Virtual Try-On, Background Management. 33 CLI commands via a2e.sh. All models FREE for max-users.

元数据

Slug a2e-image

版本 2.1.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 5

常见问题

a2e.ai Full Platform 是什么？

a2e.ai full API: Image Gen (Text2Image, NanoBanana, GPT Image, Flux 2), Video Gen (Image2Video with LoRA/FLF2V support, Video2Video, Kling 3.0, Wan 2.6, Sora... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 89 次。

如何安装 a2e.ai Full Platform？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install a2e-image」即可一键安装，无需额外配置。

a2e.ai Full Platform 是免费的吗？

是的，a2e.ai Full Platform 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

a2e.ai Full Platform 支持哪些平台？

a2e.ai Full Platform 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了 a2e.ai Full Platform？

由 DanielGrobelny（@danielgrobelny）开发并维护，当前版本 v2.1.0。

a2e.ai Full Platform