Runnable example code for the APIMODELS unified API — one API key and one OpenAI‑compatible endpoint to call image, video, LLM, audio and embedding models (GPT‑Image, Gemini, Grok, Seedance, Kling, VEO, Claude, ElevenLabs and more), at up to 60–95% lower cost than calling each provider directly.
中文说明见 README.zh-CN.md
These examples use only the public API. There are no secrets in this repo — you bring your own API key, which you generate yourself on our website.
- Sign in at apimodels.app with Google or Alipay. Signing up with Google on a personal mailbox (such as Gmail or Outlook) or with Alipay gets $0.10 of trial credit, once per person.
- Open Console → API Keys and create a key.
- Export it as an environment variable (every example reads this):
export APIMODELS_API_KEY="sk-your-key-here"Never hard‑code your key or commit it to git. Copy .env.example to .env for local use.
| Base URL | https://api.apimodels.app/v1 |
| Auth header | Authorization: Bearer $APIMODELS_API_KEY |
| LLM/chat | OpenAI‑compatible — point any OpenAI SDK at the base URL above |
# Image — synchronous because `size` is WxH and there is no callback_url.
# Returns the OpenAI shape: { "created": ..., "data": [ { "url": "https://..." } ] }
curl -X POST https://api.apimodels.app/v1/images/generations \
-H "Authorization: Bearer $APIMODELS_API_KEY" \
-H "Content-Type: application/json" \
-d '{ "model": "gpt-image-2-lite", "prompt": "a red panda barista, studio light", "size": "1024x1024", "response_format": "url" }'The trial credit is $0.10. Every example in this repo except voice cloning fits inside that on its own, but not all of them back to back: the video clip alone is $0.088. Run the video example first if you want to see it on trial credit; chat, embeddings and text‑to‑speech still fit after it.
| Example | Model | Cost per run | Fits in $0.10? |
|---|---|---|---|
chat-* |
claude-sonnet-4-6 |
< $0.001 (token‑based) | yes |
embeddings.py |
text-embedding-3-small |
fractions of a cent | yes |
text-to-speech.* |
eleven-tts-flash |
~$0.004 (min charge) | yes |
transcribe.* |
nemotron-3.5-transcribe |
$0.0025 per audio minute | yes (needs your own audio URL) |
image-generation.* |
gpt-image-2-lite @ 1024×1024 |
$0.008 | yes |
image-edit.sh |
gpt-image-2-lite @ 1536×1024 |
$0.015 | yes (needs your own image URL) |
video-generation.* |
veo-3.1-fast-fhd |
$0.088 flat | yes — on its own, run it first |
voice-clone.sh |
MiniMax voice cloning | $15.80 once per voice | no — top up first |
Two things worth knowing before you spend:
- Video is the expensive modality.
veo-3.1-fast-fhdis $0.088 for one 8‑second 1080p clip (its 720p version,veo-3.1-fast-hd, is $0.08), so it fits in the $0.10 trial credit. Many other video models cost more than the whole trial credit —grok-video-3is $0.24 for its 6‑second minimum,seedance-2.0-fast$0.28 for a 4‑second 480p clip — and will return402 Insufficient credits. - Voice cloning is a flat $15.80, charged once when the clone succeeds. That is
the upstream custom‑voice fee at cost, not a per‑second rate. Read the warning at
the top of
curl/voice-clone.shbefore running it.
Failed generations are not charged. Live prices: apimodels.app/pricing.
| Folder | Language | Examples |
|---|---|---|
curl/ |
Shell / cURL | image, image edit, video (+poll), chat, text‑to‑speech, speech‑to‑text, voice clone, balance |
python/ |
Python 3 | image, video, chat (OpenAI SDK), embeddings, TTS, speech‑to‑text, balance |
node/ |
Node.js (ESM) | image, video, chat (OpenAI SDK), TTS, speech‑to‑text, balance |
| Capability | Method & path | Notes |
|---|---|---|
| Image generation / edit | POST /images/generations |
Async by default (data.taskId → poll GET /images/generations?task_id=). Pass a WxH size and no callback_url to get a synchronous OpenAI‑shaped reply instead. Pass image_url / image_urls to edit. |
| Video generation | POST /video/generations → GET /video/generations?task_id= |
Async; poll for the result |
| Chat / LLM | POST /chat/completions |
OpenAI‑compatible |
| Embeddings | POST /embeddings |
OpenAI‑compatible |
| Text‑to‑speech | POST /tts/stream |
Streams audio bytes. This is where the eleven-tts-* models live — they are not accepted by /audio/generations. |
| Other audio (music, dialogue, dubbing, MiniMax TTS) | POST /audio/generations → GET /audio/generations?task_id= |
suno-v5, eleven-dialogue, eleven-isolator, eleven-dubbing, minimax-speech-*, kling-* |
| Speech‑to‑text (transcription) | POST /audio/generations → GET /audio/generations?task_id= |
nemotron-3.5-transcribe. Send audio_url; optional timestamps or target_language (not both). The result is text: data.resultJson is a JSON string — parse it and read text and transcript_url (the full JSON, with word timestamps and speakers when timestamps is true). $0.0025 per audio minute; up to 60 minutes / 200 MB per file. |
| Account balance | GET /balance |
Your remaining credits (USD); free, read‑only |
| Voices & cloning | GET /minimax/voices, POST /minimax/voice/clone |
List, clone or design custom voices (cloning is $15.80 per voice) |
| Model list | GET /models |
OpenAI‑compatible, no API key needed — the authoritative list of ids you can call |
Full parameter reference: apimodels.app/docs.
Every id below was checked against the live API. When in doubt,
curl https://api.apimodels.app/v1/models is the source of truth.
- Image —
gpt-image-2-lite,gpt-image-2,gemini-3-pro-image,grok-imagine-image,doubao-seedream-4-5-251128,kling-image-o1 - Video —
veo-3.1-fast-fhd($0.088 per 8‑second 1080p clip),grok-video-3,seedance-2.0-fast,seedance-2.0,kling-v2-6,kling-4.0,kling-4.0-flash(720p, 3–20 seconds, $0.094 per second; same request fields askling-4.0),minimax-h3,veo-3.1 - LLM — Claude, GPT, Gemini, DeepSeek, Qwen, GLM families (OpenAI‑compatible names — see docs)
- Audio —
eleven-tts-flash,eleven-tts-turbo,eleven-tts-multilingual,eleven-tts-v3(all on/tts/stream);minimax-speech-02-turbo,suno-v5(on/audio/generations);nemotron-3.5-transcribe(speech‑to‑text, on/audio/generations)
Task‑based endpoints return:
{ "code": 200, "msg": "success", "data": { /* ... */ } }Async tasks return data.taskId; poll the matching GET ...?task_id=<id> until
data.state is completed (then read data.resultUrls) or failed.
Speech‑to‑text is the one task whose result is text rather than a file: on a
completed nemotron-3.5-transcribe task, data.resultJson is a JSON string
({"task", "text", "duration", "transcript_url"}) — parse it once, as
transcribe.* does. data.resultUrls[0] is the same transcript_url.
The OpenAI‑compatible paths (/chat/completions, /embeddings, /models, and
/images/generations in synchronous mode) return OpenAI's own shapes instead.
Generated files are kept for 30 days, then deleted. Download anything you want to keep — the result URL stops working after that.
MIT — use these examples freely in your own projects.
Curated prompts, each shown next to the clip or image it actually produced:
- Stunning MiniMax H3 prompts — 222 video prompts (repo) · browse with video
- GPT Image 2 prompts — 955 image prompts, browsable by category
- Website: https://apimodels.app
- API Docs: https://apimodels.app/docs
- Pricing: https://apimodels.app/pricing
- Console (create API keys): https://apimodels.app/console/api-keys