deAPI AI Media Skill
Adds YouTube transcription, image/video generation, AI voice (TTS) and OCR to Claude Code and other coding assistants, with inference up to 20x cheaper.
Commands authenticate via DEAPI_API_KEY env var; no malicious behavior or covert exfiltration; data flows (uploads to api.deapi.ai) are visible in docs. However, least privilege and confirmation design are missing: /transcribe downloads external content and OCR/remove-bg upload user files to a third-party API with no explicit data-handling/retention disclosure and no rollback mechanism. Deducted for incomplete confirmation, isolation and data-flow disclosure.
Consistent submit→poll(10s)→fetch pattern across commands, error tables covering 401/429/500 and common input issues, with a clear stop-on-failure rule. But there are zero tests or execution evidence, and $ARGUMENTS is interpolated directly into JSON strings—inputs containing quotes would break requests and constitute an injection hazard. Deducted for unverified key paths and fragile interpolation.
Command table is clear and scenarios concrete (transcribe/image/TTS/OCR etc.), parameters well documented. Capability boundaries and non-fit ranges are thin; SKILL.md references ../docs/api-reference.md which breaks when the skill folder is copied standalone into ~/.claude/skills/; core function depends entirely on overseas services (api.deapi.ai, deapi.ai) with no note on mainland-China reachability. Deducted for weak boundary and environment-fit evidence.
Well-layered docs (SKILL.md routing + on-demand command docs), clear MIT license, complete install instructions. But the repo README states it is no longer actively developed (superseded by deapi-ai/skills v2), and there is no versioning or changelog; ownership/update path unclear. Deducted for missing version governance and maintenance commitment.
Broad command coverage and uniform pattern, results in principle directly downloadable. But the '20x cheaper' value claims are unverified, there are no representative output examples or comparative data, and direct usability is unconfirmed. Deducted for limited benefit evidence; statically only the happy-path structure can be endorsed.
Only author claims and marketing pricing statements; no tests, no CI evidence, no third-party execution records; model names and endpoints cannot be cross-verified from within the repository. Deducted for lack of auditable primary material and no fact/claim separation.
- Static review only; no commands were executed and usability is unreproduced.
- $ARGUMENTS is interpolated directly into JSON; inputs containing quotes or special characters may break requests—escape inputs.
- Core functions depend entirely on overseas services (api.deapi.ai etc.); mainland-China reachability is unverified and may fail.
- The repository is declared no longer actively developed; consider the v2 successor at deapi-ai/skills.
- The ../docs/api-reference.md link in SKILL.md breaks when the skill folder is installed standalone.
- OCR/background-removal commands upload user files to a third-party service; avoid with sensitive content.
What does this skill do, and when should you use it?
This is an AI media generation skill installed into Claude Code's skill directory, backed by the deAPI decentralized GPU network. Once installed, you can ask Claude in natural language to transcribe videos, generate images, convert text to speech, or extract text from images, and Claude picks the right command automatically; slash commands work too. It also supports Cursor and Windsurf, and claims up to 20x cheaper inference than proprietary APIs like OpenAI, with a free $5 credit on signup. Note that the README marks this as the v1 repository, no longer actively developed in favor of the successor skill at deapi-ai/skills targeting the deAPI v2 API.
The skill exposes 11 slash commands that call deAPI's async API (submit job, get request_id, poll every 10s, fetch result from result_url): /transcribe transcribes YouTube, Twitch, Kick, X videos or audio files; /generate-image creates images from text (Flux model); /generate-audio converts text to speech with selectable voices (e.g. am_adam); /generate-video creates video from text or animates images; /ocr extracts text from images; /remove-bg removes image backgrounds; /upscale increases image resolution 2x/4x; /transform-image applies style transfer; /embed generates text embeddings; /deapi-setup configures webhook/websocket result delivery for server apps; /deapi-balance checks account balance and credits.
- Content creators who need a YouTube video transcript plus a summary of key points from Claude
- Indie developers generating hero images for landing pages in a minimalist tech-startup style
- Bloggers converting long posts into natural speech with a professional male voice
- Anyone extracting all text from screenshots or scanned documents via OCR
- Developers building semantic search who need text embeddings
- Server-side developers configuring webhooks to receive async job results
What are this skill's strengths and limitations?
- Bundles a dozen media capabilities (transcription, image, speech, video, OCR, background removal, upscaling, embeddings) in one skill
- Natural language triggering after install — Claude routes to the right command automatically
- Claimed pricing up to 20x cheaper than OpenAI/Replicate (transcription $0.021/hour, images from $0.002/image)
- Free $5 credit on signup with no credit card required, low barrier to trying it
- Works across Claude Code, Cursor, Windsurf and Continue.dev
- The README explicitly labels this v1 repo as no longer actively developed, pointing users to the deapi-ai/skills successor
- Hard dependency on the third-party deAPI service and a paid API key; costs accrue once the free credit runs out
- Async polling every 10 seconds may be slow for long jobs
- Cursor/Windsurf get commands only, without the skill's automatic suggestions
- No test suite or independent benchmarks in the source; actual pricing and quality should be verified yourself
How do you install this skill?
1) Get an API key at deapi.ai (free $5 credit); 2) Set the environment variable: export DEAPI_API_KEY="your_key_here" (add to ~/.bashrc or ~/.zshrc); 3) Full skill install: git clone https://github.com/deapi-ai/claude-code-skills.git && cp -r claude-code-skills/deapi ~/.claude/skills/; 4) Restart Claude Code. Alternatively install individual commands only: cp claude-code-skills/deapi/commands/transcribe.md ~/.claude/commands/. For Cursor, copy commands into .cursor/commands/; for Windsurf, into .windsurf/workflows/.
How do you use this skill?
After installing, describe what you want in natural language, e.g. "Transcribe this YouTube video: https://youtube.com/watch?v=dQw4w9WgXcQ" and Claude automatically invokes /transcribe. You can also type commands directly, e.g. /generate-image a sunset over mountains or /generate-audio "Hello world" --voice am_adam. All jobs are async: submit to get a request_id, poll every 10 seconds, then fetch from result_url when done. Detailed API parameters are in docs/api-reference.md; the source does not document the full parameter tables for each command, so consult the per-command docs in the repo.
How does this skill compare with similar options?
The README positions itself against proprietary cloud APIs like OpenAI and Replicate, claiming up to 20x cost savings; the vendor also maintains a successor skill at deapi-ai/skills following the agentskills.io open standard with broader agent runtime support, which new users may prefer.