🖼

AI Image Generation API

Seedream · Hunyuan · Vidu · WAND Vega · CogView

Ten image models from five providers on one OpenAI-compatible endpoint - from a genuinely free tier to 4K, layer-output and transparent-background generation. From $0.025 per image.

🖼 AI Image Generation models & pricing

Ten image models across ByteDance Seedream, Tencent Hunyuan, Vidu, WAND Vega and Zhipu CogView. Prototype free, then scale to 4K, layer output and transparent backgrounds from $0.025 per image.

Model	Price	Strength
cogview-3-flash	FREE	Zhipu CogView, free unlimited tier	FREE
doubao-seedream-5.0	$0.027 / image	ByteDance Seedream, photorealism + text rendering
doubao-seedream-5.0-pro	$0.034 / image	Max detail, commercial hero images
hy-image-v3	~$0.031 / image	Tencent Hunyuan, strong prompt adherence	NEW
vidu-image-q2	from ~$0.029 / image	Vidu, 1080p-4K, 0-7 reference images	NEW
seedream-image-v5.0-lite	~$0.034 / image	Seedream, 2K-4K, group images, web-search tool	NEW
seedream-image-v5.0-pro	from ~$0.046 / image	Seedream, layer decomposition, transparent bg	NEW
wand-vega-image-lite	from ~$0.025 / image	Tencent WAND Vega, 1K-4K, cheapest	NEW
wand-vega-image-flash	from ~$0.069 / image	WAND Vega, 1K-4K, 0-6 reference images	NEW
wand-vega-image-pro	from ~$0.146 / image	WAND Vega, layer output, 0-6 references	NEW

Why run these models on NovAI?

Models

10 image models

Cheapest

FREE (CogView)

From

$0.025 / image

Max res

Ref images

Up to 7

Platform fee

One OpenAI-compatible endpoint (/v1/images/generations) returning {created, data:[{url}]}
Free unlimited CogView 3 Flash tier for prototyping (watermarked until any top-up)
Up to 4K resolution and 0-7 reference images for image-to-image editing
Pro features: layer decomposition and transparent backgrounds (Seedream Pro, Vega Pro)
Async upstream models (Vidu, Vega) are auto-polled internally - you still get a synchronous response

🚀 API usage

cURL — synchronous image generation

curl https://aiapi-pro.com/v1/images/generations \
  -H "Authorization: Bearer $NOVAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "hy-image-v3",
    "prompt": "a cute orange cat on a windowsill, soft sunlight",
    "size": "1024x1024",
    "n": 1
  }'

Python (OpenAI SDK)

from openai import OpenAI
client = OpenAI(base_url="https://aiapi-pro.com/v1", api_key="YOUR_NOVAI_API_KEY")
resp = client.images.generate(
    model="hy-image-v3",
    prompt="a cute orange cat on a windowsill, soft sunlight",
    size="1024x1024", n=1,
)
print(resp.data[0].url)

All image models use POST /v1/images/generations and return an OpenAI-compatible {created, data:[{url}]} response. Async upstream models (Vidu, WAND Vega) are polled to completion internally (up to ~180s) before the URL is returned. Charge = usage.total_tokens / 1000 × token_price; per-image figures are base-resolution references that scale with resolution & reference-image count.

Best use cases

E-commerce product photography
Marketing and ad creative
Illustrations, cover art and concept design
Image editing with reference photos
High-volume content pipelines

Frequently asked questions

Which image model is cheapest?

cogview-3-flash is completely free (unlimited, watermarked until any top-up). Paid models start at wand-vega-image-lite (~$0.025) and doubao-seedream-5.0 ($0.027) per image.

Is the image API OpenAI-compatible?

Yes. Every image model uses POST /v1/images/generations and returns the standard {created, data:[{url}]} shape, so client.images.generate(...) works with the OpenAI SDK.

Do you support image-to-image and reference images?

Yes - Vidu (0-7 references), WAND Vega (0-6) and Hunyuan Image 3.0 (0-3) accept reference images for editing and style transfer.

What about resolution and special outputs?

Models support 1K to 4K. Seedream 5.0 Pro adds layer decomposition and transparent backgrounds; WAND Vega Pro adds layer output.