Image Generation & Editing

HJToken supports the official OpenAI gpt-image models for text-to-image generation and image editing. The API matches OpenAI's exactly — point your existing SDK at a new base_url and it works.

Supported models

Model	Position	Notes
gpt-image-2	Flagship	Best quality and prompt adherence — use for final deliverables
gpt-image-1.5	Previous flagship	Stable, slightly cheaper
gpt-image-1	Legacy	OpenAI deprecates it on 2026-10-23 — don't use for new work
gpt-image-1-mini	Budget	Cheapest; ideal for bulk drafts and prompt iteration

The authoritative list is the model pricing page, which syncs automatically with upstream.

Billing: per output token, not per image

This is the part worth understanding up front, because it differs from many platforms:

gpt-image models are billed by token. A request costs roughly:

prompt tokens × input rate
+ reference image tokens × image input rate   (editing / reference images only)
+ generated image tokens × image output rate

Higher quality and larger size produce more output tokens, so they cost more. For the same 1024×1024 image:

Quality	Approx. output tokens	Relative cost
low	a few hundred	lowest
medium	a few thousand	moderate
high	4,000+	highest

So while iterating on prompts, use low quality with gpt-image-1-mini and only switch to high quality for the final render — that alone removes most of the cost.

Per-model image output rates are in the "Image output (/1M)" column on the model pricing page. Actual token counts and charges are itemized under usage.

Option 1: generate directly in Chat (no code)

1. Open Chat
2. Pick a gpt-image-* model in the model selector
3. Switch the mode to Image (or Edit and upload a source image)
4. Choose a quality tier (low / medium / high), type your description, send

Download the result directly. Charges are identical to the API path and come from the same balance.

Option 2: call the API

Base URL: https://api.hjtoken.cn

Text-to-image — /v1/images/generations

curl https://api.hjtoken.cn/v1/images/generations \
  -H "Authorization: Bearer $HJTOKEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A shiba inu wearing an astronaut helmet, cyberpunk neon backdrop, cinematic lighting",
    "size": "1024x1024",
    "quality": "medium",
    "n": 1
  }'

data[].b64_json in the response is the base64-encoded image — decode it to a file:

cat response.json | jq -r '.data[0].b64_json' | base64 -d > out.png

Image editing — /v1/images/edits

Editing uses multipart/form-data with the source image as a file upload:

curl https://api.hjtoken.cn/v1/images/edits \
  -H "Authorization: Bearer $HJTOKEN_API_KEY" \
  -F model="gpt-image-2" \
  -F image="@input.png" \
  -F prompt="Replace the background with a snowy mountain at sunset, keep the subject unchanged" \
  -F size="1024x1024" \
  -F quality="medium"

With the OpenAI SDK

Change only base_url and api_key; everything else follows OpenAI's docs:

from openai import OpenAI

client = OpenAI(
    api_key="your HJToken API key",
    base_url="https://api.hjtoken.cn/v1",
)

result = client.images.generate(
    model="gpt-image-2",
    prompt="Minimal line-art illustration of mountain ranges, off-white background",
    size="1024x1024",
    quality="low",
)

import base64
with open("out.png", "wb") as f:
    f.write(base64.b64decode(result.data[0].b64_json))

Common parameters

Parameter	Values	Notes
prompt	text	Image description. The more specific (subject, style, composition, lighting), the more consistent the result
size	1024x1024, 1024x1536, 1536x1024, …	Larger sizes consume more output tokens
quality	low / medium / high	The parameter with the largest cost impact
n	integer	How many images per request. Cost scales per image

FAQ

Why base64 instead of a URL? The official gpt-image API returns base64 and does not host the result. It also means the image never touches third-party storage.

Does HJToken store my images? No. The gateway only forwards the request and records billing metadata (token counts and cost) — image content is not retained.

Can I generate video? Not yet. Only image generation is available today.

Why was this image more/less expensive than I expected? Check the output token count for that request under usage — cost is directly proportional to it. Going from low to high quality can be a 10×+ difference in tokens.

I get a "model not available" error. Your current plan or group doesn't include that image model. Check what you can reach under available channels, or contact us to enable it.