--- name: wavespeed description: | Generate images, video, audio and avatars (700+ models) or call OpenAI-compatible LLMs (290+ models — Claude, GPT, Gemini, DeepSeek, Llama) through the WaveSpeedAI inference platform. Use whenever the user asks to create AI media, run a `wavespeed-ai/*` model, upload a media asset for inference, list available models, check account balance, or call an LLM via the WaveSpeed proxy. --- # WaveSpeedAI — generic skill (any agent that follows the agentskills.io spec) This file is intentionally portable. Copy it into whichever skill directory your host agent reads from. ## Setup ```bash export WAVESPEED_API_KEY="ws_..." # https://wavespeed.ai/accesskey ``` CLI shim: `~/.local/bin/wavespeed-cli` (run `bash install.sh` from the repo). ## Trigger conditions - User asks for AI media generation (image / video / audio / avatar). - User mentions a `wavespeed-ai/*` uuid. - User wants to upload a local file as input for image-to-X. - User wants to call Claude / GPT / Gemini / Llama via the WaveSpeed LLM proxy. - User asks for the WaveSpeed model catalog or balance. ## Surface ```bash wavespeed-cli run '' # submit + poll until terminal wavespeed-cli submit '' # submit only; print task id wavespeed-cli result # fetch result wavespeed-cli cancel # cancel/delete wavespeed-cli upload # upload local asset wavespeed-cli models [--names-only|--filter] # catalog wavespeed-cli balance # account balance wavespeed-cli llm '' # OpenAI-compatible chat wavespeed-cli verify-webhook # HMAC-SHA256 webhook verify ``` ## Examples ```bash # Text → image wavespeed-cli run wavespeed-ai/z-image/turbo '{"prompt":"Cat in tuxedo"}' # Image → video URL=$(wavespeed-cli upload ./input.png) wavespeed-cli run wavespeed-ai/seedance-v2 '{"image":"'"$URL"'","prompt":"camera dolly"}' # Text → video w/ audio wavespeed-cli run wavespeed-ai/veo-3 '{"prompt":"thunderstorm at dusk, cinematic"}' # LLM wavespeed-cli llm anthropic/claude-opus-4.6 "Summarize in one sentence." ``` ## Python SDK ```python import wavespeed out = wavespeed.run("wavespeed-ai/z-image/turbo", {"prompt": "Cat"}) print(out["outputs"][0]) url = wavespeed.upload("/path/to/image.png") ``` Use the dedicated venv: `~/.local/share/wavespeed-skill/venv/bin/python script.py`. ## Errors - `401` → key missing or revoked. - `400` → payload doesn't match the model schema. - `402` → insufficient balance / tier locked. - `404` → task or model not found. - `429` → rate-limited; back off. - `5xx` → upstream error; retry idempotent submits. ## Refs - Docs: https://wavespeed.ai/docs - Skill repo: https://github.com/wesleysimplicio/WaveSpeedAI-Skills - Python SDK: https://github.com/WaveSpeedAI/wavespeed-python - API key: https://wavespeed.ai/accesskey