novita-ai

Original：🇺🇸 English

Translated

Novita AI: LLM, Image Generation & Editing, Video Generation, Audio (TTS/ASR), and GPU Cloud. Use this skill whenever the user wants to call Novita AI APIs — chat with LLMs (DeepSeek, Llama, Qwen), generate images (FLUX, Stable Diffusion, Seedream, Hunyuan Image), edit images (remove background, upscale, inpainting, img2img, outpainting, reimagine, merge face, replace background, remove text), generate videos (Kling, Wan, Hunyuan, Minimax Hailuo, Vidu, PixVerse, Seedance), do text-to-speech or speech-to-text (MiniMax TTS, GLM TTS, Fish Audio, ASR, voice cloning), run OpenAI-compatible batch jobs, manage GPU cloud instances and serverless endpoints, or check account balance and billing. Also trigger when the user mentions novita.ai, Novita AI, Novita API key, or wants to use any Novita platform service — even if they just say "generate an image" or "run an LLM" and Novita is available as a provider.

2installs

Sourcenovitalabs/novita-skills

Added on2026-03-25

NPX Install

npx skill4agent add novitalabs/novita-skills novita-ai

SKILL.md Content

View Translation Comparison →

Novita AI

Access 200+ AI models through a unified API — LLM chat and embeddings (DeepSeek, Llama, Qwen), image generation and editing (FLUX, Stable Diffusion, Seedream), video generation (Kling, Wan, Hailuo, Vidu), text-to-speech, speech recognition, and GPU cloud infrastructure.

OpenAI-compatible LLM API works as a drop-in replacement with any OpenAI SDK
30+ image endpoints covering generation, editing, upscaling, background removal, face merging, and more
Video generation from 7+ providers including Kling, Wan, Minimax Hailuo, Vidu, and Seedance
Full GPU cloud management — instances, templates, storage, and serverless endpoints

Setup

Get an API key at novita.ai/settings/key-management
Set the environment variable:
```
export NOVITA_API_KEY=your_key
```
All requests use
```
Authorization: Bearer $NOVITA_API_KEY
```
header
Base URL:
```
https://api.novita.ai
```
Full documentation: novita.ai/docs/api-reference

Services

Service	Use When	Endpoints	Mode
LLM	Chat, completion, embeddings, reranking	`/openai/v1/*`	Sync / Stream
Image Generation	Text-to-image, image-to-image	FLUX, SD, Seedream, Qwen, Hunyuan, GLM	Sync / Async
Image Editing	Remove BG, upscale, inpaint, outpaint, cleanup, reimagine, merge face	`/v3/*`	Sync / Async
Video Generation	Text-to-video, image-to-video	Kling, Wan, Hunyuan, Hailuo, Vidu, PixVerse, Seedance	Async
Audio	TTS, ASR, voice cloning	MiniMax, GLM, Fish Audio	Sync
Batch	Bulk LLM processing	`/openai/v1/batches`	Async
GPU Cloud	Instances, templates, storage, serverless	`/gpu-instance/openapi/v1/*`	Sync
Account	Balance, billing	`/openapi/v1/billing/*`	Sync

LLM (OpenAI-Compatible)

The LLM API is a drop-in replacement for the OpenAI API. Use the standard OpenAI SDK — just change the base URL to

https://api.novita.ai/openai

python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.novita.ai/openai",
    api_key="YOUR_NOVITA_API_KEY",
)

response = client.chat.completions.create(
    model="deepseek/deepseek-v3-0324",
    messages=[{"role": "user", "content": "Hello"}],
    max_tokens=512,
)

Method	Path	Purpose
POST	`/openai/v1/chat/completions`	Chat completion (streaming supported)
POST	`/openai/v1/completions`	Text completion
POST	`/openai/v1/embeddings`	Text embeddings
POST	`/openai/v1/rerank`	Document reranking
GET	`/openai/v1/models`	List available models

Key models:

deepseek/deepseek-v3-0324

deepseek/deepseek-r1

meta-llama/llama-3.1-70b-instruct

Qwen/Qwen2.5-72B-Instruct

. Embedding:

baai/bge-m3

. Rerank:

baai/bge-reranker-v2-m3

. Full list via

/models

Features: vision (multimodal), reasoning, function calling, structured outputs, prompt caching, batch API.

For full parameters and examples → references/llm-api.md

Image Generation

Endpoint	Path	Mode	Description
FLUX.1 Schnell	`POST /v3beta/flux-1-schnell`	Sync	Fast text-to-image, cheapest option
FLUX Kontext Dev/Pro/Max	`POST /v3/async/flux-1-kontext-*`	Async	Advanced generation with image editing
FLUX 2 Dev/Flex/Pro	`POST /v3/async/flux-2-*`	Async	Latest FLUX models
Stable Diffusion txt2img	`POST /v3/async/txt2img`	Async	Full SD control with LoRAs, samplers
Stable Diffusion img2img	`POST /v3/async/img2img`	Async	Image-to-image transformation
Seedream 3.0 / 4.0 / 4.5 / 5.0-lite	Async	Async	ByteDance image models
Qwen Image / Hunyuan Image 3 / GLM Image	Async	Async	Additional providers

For full parameters and examples → references/image-api.md

Image Editing

Endpoint	Path	Mode	Input
Remove Background	`POST /v3/remove-background`	Sync	Base64 image
Replace Background	`POST /v3/replace-background`	Sync	Base64 image + text prompt
Reimagine	`POST /v3/reimagine`	Sync	Base64 image
Image to Prompt	`POST /v3/img2prompt`	Sync	Base64 image → returns text
Remove Text	`POST /v3/remove-text`	Sync	Base64 image
Cleanup (erase)	`POST /v3/cleanup`	Sync	Base64 image + base64 mask
Outpainting	`POST /v3/outpainting`	Sync	Base64 image + prompt + dimensions
Merge Face	`POST /v3/merge-face`	Sync	Base64 face + base64 target
Upscale	`POST /v3/upscale`	Sync	Base64 image
Inpainting	`POST /v3/async/inpainting`	Async	Base64 image + base64 mask + prompt

All image inputs use base64-encoded local files. For full parameters → references/image-api.md

Video Generation

All video endpoints are async — they return a task_id for polling. The Unified Video API provides a single endpoint for all models.

Endpoint	Path	Description
Unified Video API	`POST /v3/video/create`	Single endpoint for all video models
SD Text-to-Video	`POST /v3/async/txt2video`	Legacy Stable Diffusion video
SD Image-to-Video	`POST /v3/async/img2video`	SVD/SVD-XT models
Hunyuan Video Fast	`POST /v3/async/hunyuan-video-fast`	Cost-effective text-to-video

Available Video Models

Model	Capabilities	Notable Features
Kling v2.1 / v2.5 / v2.6	T2V, I2V, ref2v, video-edit	Motion control, camera control
Wan 2.5 / 2.6	T2V, I2V	Fast preview modes
Minimax Hailuo 02 / 2.3	T2V, I2V	Fast I2V variant
Hunyuan Video Fast	T2V	Cost-effective
Vidu Q1 / Q2 / Q3	T2V, I2V, startend2v, ref2v	Multi-frame, templates
PixVerse v4.5	T2V, I2V	—
Seedance v1 / v1.5	T2V, I2V	Lite and Pro variants

For full parameters → references/video-api.md

Audio

Endpoint	Path	Mode	Use When
MiniMax Speech 02 HD	`POST /v3/minimax-speech-02-hd`	Sync	English TTS, high quality, 17 voices
GLM TTS	`POST /v3/glm-tts`	Sync	Chinese TTS, low latency, 7 voices
GLM ASR	`POST /v3/glm-asr`	Sync	Speech-to-text transcription
MiniMax Voice Cloning	`POST /v3/minimax-voice-cloning`	Sync	Clone a voice from audio
Fish Audio TTS	Async	Async	Custom voice TTS

MiniMax voices: Wise_Woman, Calm_Woman, Friendly_Person, Deep_Voice_Man, Inspirational_girl, Casual_Guy, Lively_Girl, Patient_Man, Young_Knight, Lovely_Girl, Sweet_Girl_2, Elegant_Man

GLM voices: tongtong, chuichui, xiaochen, jam, kazi, douji, luodo

Emotion control (MiniMax): happy, sad, angry, fearful, disgusted, surprised, neutral

For full parameters → references/audio-api.md

Async Task Polling

Async endpoints (SD images, video, some audio) return a task_id. Poll for results:

GET /v3/async/task-result

with query parameter

task_id

Status lifecycle: TASK_STATUS_QUEUED → TASK_STATUS_PROCESSING → TASK_STATUS_SUCCEED or TASK_STATUS_FAILED

Poll every 3-5 seconds. On success, results contain download URLs in

images[]

videos[]

, or

audios[]

arrays. You can also configure a webhook callback in the original request for async notification.

Batch Processing (OpenAI-Compatible)

For bulk LLM jobs using the OpenAI-compatible batch API at

/openai/v1

Upload a JSONL file via
```
POST /openai/v1/files
```
Create a batch via
```
POST /openai/v1/batches
```
with the file ID
Poll batch status via
```
GET /openai/v1/batches/{batch_id}
```

Download results via

GET /openai/v1/files/{output_file_id}/content

For full parameters → references/llm-api.md

GPU Cloud

Base:

/gpu-instance/openapi/v1

Operation	Method	Path
List GPU products	GET	`/products`
List CPU products	GET	`/cpu/products`
List clusters	GET	`/clusters`
Create instance	POST	`/gpu/instance/create`
List instances	GET	`/gpu/instance/list`
Get instance	GET	`/gpu/instance/get`
Start / Stop / Restart	POST	`/gpu/instance/{action}`
Delete instance	POST	`/gpu/instance/delete`

GPU creation costs real money — always check

/products

for pricing first.

Also supports: Templates (create/list/get/update/delete), Network Storage (create/list/delete), Serverless Endpoints (create/list/get/update/delete).

For full parameters → references/gpu-api.md

Account and Billing

Operation	Method	Path
Get balance	GET	`/openapi/v1/billing/balance/detail`
Monthly bill	GET	`/openapi/v1/billing/monthly/bill`
Usage-based billing	GET	`/openapi/v1/billing/bill/list`
Fixed-term billing	GET	`/openapi/v1/billing/fixed-term/bill`

Balance amounts are in units of 0.0001 USD (divide by 10000 for dollars).

Decision Guide

Which image endpoint?

Intent	Endpoint	Notes
Text to image (fast)	FLUX.1 Schnell	Sync, cheapest
Text to image (quality)	FLUX Kontext, Seedream, SD	Async, more control
Image + text to new image	img2img	Style transfer
Edit region with prompt	Inpainting	Needs mask
Erase region	Cleanup	Sync, needs mask
Extend canvas	Outpainting	Sync
Remove background	remove-background	Sync
New background	replace-background	Sync
Remove text overlay	remove-text	Sync
Describe image	img2prompt	Returns text
Enlarge and enhance	Upscale	Sync
Swap face	merge-face	Sync
Restyle completely	Reimagine	Sync

Which TTS?

English, multilingual, high quality: MiniMax Speech 02 HD
Chinese, low latency: GLM TTS
Custom voice: Fish Audio or MiniMax Voice Cloning

Which video model?

General purpose: Kling v2.5 or Hailuo 02
Fast and cheap: Hunyuan Video Fast, Wan 2.5 Preview
High quality: Kling v2.6 Pro, Vidu Q3
Image-to-video: Kling I2V, Wan I2V, Hailuo fast-I2V

Security

Never hardcode API keys in code or commit them to version control. Use environment variables or secret managers.
Endpoints accepting media inputs should only receive content from trusted, verified sources.
When using async webhook callbacks, ensure your callback server validates the request origin.
Enable NSFW detection for user-facing applications.

Error Handling

Code	HTTP Status	Meaning
INVALID_API_KEY	403	Bad or missing API key
RATE_LIMIT_EXCEEDED	429	Too many requests
BILLING_BALANCE_NOT_ENOUGH	400	Insufficient balance
MODEL_NOT_FOUND	404	Invalid model name
INVALID_REQUEST_BODY	400	Malformed request
SERVICE_NOT_AVAILABLE	503	Service temporarily down

API References

For detailed endpoint parameters, request and response schemas, and code examples:

LLM API: references/llm-api.md — Chat, embeddings, rerank, function calling, structured outputs, batch
Image API: references/image-api.md — All generation and editing endpoints with full parameter specs
Video API: references/video-api.md — Unified API, model-specific parameters
Audio API: references/audio-api.md — TTS variants, ASR, voice cloning, streaming
GPU Cloud API: references/gpu-api.md — Instances, templates, storage, serverless