Skip to main content
LaoZhang API’s image lineup centers on three families, Google Nano Banana, OpenAI GPT Image, and xAI Grok Imagine, all called with one LaoZhang API key and mostly billed per call or per image. Pick by protocol and budget; the cards below give the recommended model, price, and endpoint for each family.

​
Main image model families

nb21-en-generate

Nano Banana

Google, recommended Nano Banana 2.1. $0.045 per call at 1K–4K; Gemini native or OpenAI-compatible, up to 14 references.
gpt25-en-card

GPT Image 2.5

OpenAI, Flare and Sunburst. VIP $0.03 per call, official forwarding by tokens; OpenAI Images API with edits and transparent PNG.
grok-generate

Grok Imagine 2.0

xAI’s current image model. $0.055 per returned image; OpenAI Images API, edits with up to 3 references.
Each card shows the actual output of the generate example on that model’s guide. Capabilities side by side (verified October 14, 2026): What the three families share:
  • Every model comes from the vendor’s official channel. No Google, OpenAI, or xAI account is needed; your LaoZhang API key is the only credential, and the console call log shows what each request cost.
  • On per-call and per-image models, resolution, quality, and reference count don’t change the price. GPT Image official forwarding bills by tokens, so size and quality do change its cost.
  • Whether a request with no image is charged depends on the model; each guide states the rule, and About LaoZhang API summarizes them.
  • Launches, suspensions, and recoveries are posted in Announcements; current availability is on the status page.
For other tasks, choose Seedream 5.0 Flash or Nano Banana 2 Lite for low-cost batches, and FLUX.2 or Seedream to combine many references in one JSON request. The full table follows. The table covers every current image model with its endpoint, billing, and price; legacy IDs are listed after it. Prices are LaoZhang API public pricing as of October 7, 2026. Check the model and pricing catalog or the console before a large batch.

​
Compare image models

All requests use the base URL https://api.laozhang.ai. The protocol labels mean:
  • Images API: generate with POST /v1/images/generations (JSON body) and edit with POST /v1/images/edits (multipart file upload).
  • Images generations: both tasks use POST /v1/images/generations. To edit, add reference images to the JSON body as URLs or data URLs.
  • Gemini native: POST /v1beta/models/{model}:generateContent. Use it when you need to choose the resolution or aspect ratio of a Nano Banana image.
  • Chat Completions: the OpenAI-compatible POST /v1/chat/completions endpoint.
Each billing type needs a matching token:
  • Token-billed rows (GPT Image 2.5 official forwarding and gpt-image-2 in the official-forward groups) use official API forwarding. They need a token in the OpenAI Official (formerly Sora2Official) or OpenAI Official Enterprise (formerly GPTImage2 Sora2 Enterprise) group with the Usage first (按量优先, recommended) or Usage-based (按量计费) billing mode.
  • Per-call models use a default-group token with Usage first or Per-call (按次计费). The GPT Image per-call models also work in another image group (see the console for its name).
A Usage first token can call both kinds of model. You choose the group when you create the token in token management, not in the request’s model field.

​
Official-forward token rates

gpt-image-2.5-flare, gpt-image-2.5-sunburst, and gpt-image-2 on the official-forward route are billed by actual token usage. OpenAI’s official rates are: The token’s group sets what you pay. OpenAI Official charges the same as OpenAI’s official price, and OpenAI Official Enterprise charges the official price plus 20% (a 1.2× group multiplier). The same group pricing applies to gpt-image-1, gpt-image-1.5, and gpt-image-1-mini in these groups. Token rates don’t translate into a fixed cost per image: size, quality, and reference images all change usage, so estimate costs from a few representative requests.

​
Estimate cost by volume

For per-call and per-image models, cost is the price times the number of images; size and quality don’t change it. Estimates from public pricing as of October 7, 2026, in USD:
  • The estimate assumes one image per request; Grok Imagine bills each image when one request returns several.
  • With Google Search grounding on, Nano Banana adds $0.014 per search.
  • Actual charges are in the call log, and the model catalog has current prices.

​
Choose by task

  • Generate and edit through one OpenAI-compatible API: use GPT Image 2.5. The -vip models bill per call and accept custom sizes and five quality tiers; the official-forward models keep OpenAI’s own parameters and bill by tokens.
  • Transparent PNG assets: the GPT Image 2.5 -vip models generate transparent PNGs and can remove the background from an uploaded image. To split a finished image into a background plus transparent layers, use Seedream layer decomposition on 5.0 Flash or 5.0 Pro. Both bill each output image, and Flash costs far less per image.
  • Low-cost images at high volume: use Seedream 5.0 Flash ($0.018 per call, up to 2K, strong at designs with text) or Nano Banana 2 Lite for 1K drafts. Nano Banana Standard (gemini-2.5-flash-image) costs less per call than Nano Banana 2 Lite, but Google has deprecated it and plans to shut it down on March 15, 2027 (deprecations page). LaoZhang API still serves it for now, and it can stop at any time. Migrate to Nano Banana 2.
  • 2K or 4K output from a Gemini model: use Nano Banana 2.1, which costs $0.045 per call at any size, Nano Banana 2, or Nano Banana Pro through the Gemini native endpoint.
  • Many reference images in one request: Nano Banana 2.1, Nano Banana 2, and Nano Banana Pro accept up to 14 in the Gemini contents array, Seedream up to 10, and FLUX.2 up to 8 as JSON fields.
  • Several candidates from one prompt: Grok Imagine 2.0 returns up to ten images per request and bills each returned image.
  • Existing gpt-image-2 or gpt-image-2-vip code: keep it running with the GPT Image 2 guide, and build new integrations on GPT Image 2.5.

​
Legacy and historical model IDs

These IDs still appear in the catalog, but new integrations should use the models in the table above:
  • gpt-image-2-web ($0.03 per call) is a historical web-route ID used by existing integrations, supports both generation and editing, and renders natively at up to about 2K, with 4K upscaled. It is a different route from gpt-image-2.5-web, and their parameters are not interchangeable.
  • grok-imagine-image-quality ($0.045 per image) is the earlier Grok Imagine quality model, billed per successfully returned image; see the Grok Imagine guide.
    • xAI announced its retirement from November 2, 2026. As of November 3 it is still callable on LaoZhang API.
    • The earlier grok-imagine-image was withdrawn on October 30, 2026.
    • New integrations should use grok-imagine-image-2.0.
  • flux-kontext-pro ($0.035 per call) and flux-kontext-max ($0.07 per call) are the legacy FLUX Kontext models, covered in the legacy section of the FLUX.2 guide.
  • gpt-image-1, gpt-image-1.5, and gpt-image-1-mini are earlier token-billed GPT Image models. Their rates are in the model and pricing catalog. OpenAI plans to remove gpt-image-1 from its API on October 23, 2026 and the other two on December 1, 2026; LaoZhang API stops serving each at the same time, so use GPT Image 2.5 for new integrations.

​
Frequently asked questions

​
Which model should I use for high-volume batches?

Start with a per-call model, because each request has a fixed price and batch costs are easy to predict. For new low-cost work, try Seedream 5.0 Flash (seedream-5-0-flash-260915, $0.018 per call), which outputs up to 2K and handles designs with text well, or Nano Banana 2 Lite (gemini-3.1-flash-lite-image) for 1K output. If you need the OpenAI Images API, the two GPT Image 2.5 -vip models also bill per call, and size and quality don’t change their price. Nano Banana Standard (gemini-2.5-flash-image) costs less than Nano Banana 2 Lite, but Google has deprecated it and LaoZhang API can stop serving it at any time, so don’t build new batch jobs on it. Grok Imagine 2.0 bills every returned image, so a request that returns ten images costs ten times its unit price. Official-forward models bill by tokens, and the cost per image changes with size, quality, and reference images, so estimate from a few representative requests before a large run.

​
Which model should I use to edit an existing image?

If you already use the OpenAI Images API and want to preserve as much of the source image as possible, use GPT Image 2.5 Sunburst: gpt-image-2.5-sunburst-vip for per-call billing or gpt-image-2.5-sunburst for token billing. To limit the change to a region with a mask, use official-forward GPT Image 2.5, because the -vip models don’t support mask. The web route, gpt-image-2.5-web, also edits images but can’t select Flare or Sunburst, so use one of the two Sunburst IDs when you need its edit precision. To combine several reference images, choose by the reference limits in Choose by task. Each model takes references differently, and each model guide has a complete example:
  • GPT Image and Grok Imagine upload files as multipart form data on /v1/images/edits.
  • FLUX.2 and Seedream take URLs or data: URLs in the JSON body.
  • Nano Banana takes images inside the message or the Gemini contents array.

​
Where do I confirm prices? How do per-call and token billing differ?

Before you call, check the model and pricing catalog and the console. Each request’s actual charge appears in call logs, and enterprise quotes come from the support team. Per-call models charge a fixed amount for each successful request, while official-forward models charge for the tokens actually used. The same gpt-image-2 ID is billed differently depending on the token’s group, and changing model doesn’t change that. The GPT Image 2 guide shows each route.