Flux ArtShort answer first: the GPT Image 2 API on Flux Art (an all-in-one AI visual generation workbench that aggregates 50+ image and video models under one account) runs on a points system, not a fixed price per call. Images are billed per image generated, and size and quality affect the consumption of some models — the actual charge is whatever `usage.points_charged` returns in the task response. The API base URL is `https://open-api.flux-art.ai/openapi/v1`, and the console is reachable from both https://flux-art.ai and https://flux-art.cn. The API and the web app share the same points balance — it's not two separate wallets.
What Exactly Does the GPT Image 2 API Charge For?
Let's get the billing model straight first — it determines which cost-saving moves are actually worth making later.
| Billing Dimension | Official Statement | What It Means for You |
|---|---|---|
| Billing unit | Points, not a fixed price per call | Different models and parameters consume different amounts — go by the actual response |
| Image billing | Charged per image generated; size and quality affect some models' consumption | 4K and 1K aren't the same price — parameters are cost switches |
| Video billing | Charged by generated duration; resolution, audio, and video input mode affect cost, rounded up to a 50-point minimum unit | Don't chop short videos into too many pieces — every segment gets rounded up |
| Source of truth | Whatever `usage.points_charged` shows in the task response | Don't guess — read the response value and book it |
| Failure refunds | Qualifying validation failures are refunded, logged as `usage.points_refunded` | A wrong parameter won't burn your points for nothing |
| Insufficient balance | Returns `402 insufficient_points` directly; no task is created | There's no such thing as being charged with no image delivered |
| Relationship to the web app | Shares account points, membership perks, and current discounts | No need to buy a separate subscription just to try the API |
The last two rows are the ones worth remembering. The `402` design means that when your balance runs dry, it "just refuses to work" instead of charging you half the points for half an image; shared points mean parameters you've already tuned on the web app can go straight into the API without paying twice.
How Much Does One Image Actually Cost? Can I Get a Number?
- What's documented: new users get 500 points on sign-up, good for roughly 30+ GPT Image 2 images.
- Working backward: 500 ÷ 30 ≈ roughly 16 points per image (this is a rough upper-bound estimate, because "30+" means the real number of images is at least that many, so the true per-image cost is probably lower than 16 points).
- That number only holds near the default settings. Once you push `size` to a higher resolution or bump up the quality tier, per-image consumption climbs.
As for subscription tiers, the site currently lists Free at $0, Pro at $15, Max at $35, and Ultra at $95, billed monthly or annually — annual billing saves about 47%. Pro and above unlock every feature with no throttling, and Pro/Max/Ultra support up to 4K output plus Open API Support. GPT Image 2 and the whole Nano Banana lineup also have a limited-time 50% discount. Prices and promotions can change, so check the site for the current numbers.
How I Actually Brought Down the Cost of a Batch
The problem with version one wasn't image quality — it was that it cost way more than it needed to. Going back through the workflow, I realized only part of that batch was ever going to end up as listing hero images. The rest was for internal SKU selection and group-chat pitch comparisons, where rendering in high-res was pure waste. Parameters are cost switches, and I'd run the entire batch with every switch flipped to max.
One more trap worth flagging: don't hardcode a polling loop. The account-level task-read limit is 120 requests per minute, and when I first started polling twice a second, running just three tasks in parallel was enough to hit `429`. My current approach waits two or three seconds before the first check, then backs off progressively, and honors the `Retry-After` response header whenever a 429 comes back. This doesn't save points, but it saves the time your whole batch loses to being throttled.
One more thing: when you retry, reuse the same idempotency key from the original request — that's the whole point of idempotency keys. It means a timeout or 5xx retry won't double-charge your points.
How Do You Call the API to Spend Fewer Points?
Ranked by return on effort, the top three matter the most:
- Run tiered: low tier for drafts, high tier for delivery. Most teams waste points "rendering high-res images that never go live."
- Only max out parameters when you actually need to: `size` and quality tier directly affect consumption, so don't default to full blast.
- Get idempotency keys right: reuse the original key on timeout/5xx retries to avoid double charges; use a fresh key for a genuinely new request to avoid a 409.
- Book actuals from `points_charged`: build your cost model on real values, not estimates.
- Dial in parameters on the web app first: since points are shared, whatever works on the web app carries straight over to the API, skipping the trial-and-error cost on the API side.
- Pay annually: saves about 47% versus monthly (check the site for the current number).
Find Your Situation: What to Do on Flux Art
| Your Situation | Biggest Pain Point | What to Do on Flux Art | Recommended Model |
|---|---|---|---|
| Want to nail down the per-image cost first | No published unit price | Run 20 images to calibrate, read the real `usage.points_charged` value, then multiply by volume | GPT Image 2 (`gpt-image-2`) |
| Batch hero-image runs blow the budget | The whole batch renders in high-res | Split into draft-tier + delivery-tier passes; only render selected images at the high tier | GPT Image 2, Nano Banana 2 |
| Worried a wrong parameter burns points for nothing | Trial-and-error is costly | Qualifying validation failures are refunded — check `usage.points_refunded`; insufficient balance returns 402 without creating a task | Pick per your need |
| Already have a web app subscription | Worried the API needs its own payment | No need — it shares the same points and membership perks; just create a key and call it | Pick per your need — one key calls everything |
| Integrating video | Billing rules are unclear | Charged by duration, rounded up to a 50-point minimum unit — don't chop clips into too many segments | Seedance 2.0 (`doubao-seedance-2-0-260128`) |
Where Does It Actually Save You Money Versus Subscribing to Each Vendor Directly?
It's not just money you save — it's management overhead. Subscribing to each vendor separately typically means multiple bills, multiple keys, multiple integrations, and juggling payment methods and account stability on top of it. Once everything sits under one account, it's one subscription, one key, one `POST /images/generations` endpoint — you switch models just by changing the `model` field. GPT Image 2, Nano Banana, Seedream, and Midjourney all live behind the same interface.
Whether this math is worth doing depends on your volume. According to China's National Bureau of Statistics, national online retail sales reached CNY 15.9722 trillion in 2025, up 8.6% year over year, with physical goods online retail sales at CNY 13.0923 trillion — 26.1% of total retail sales of consumer goods. With roughly a quarter of retail happening online, product images are an ongoing consumable, not a one-time expense — the higher your volume, the wider the gap grows between per-image cost and management overhead.
Flux Art is an all-in-one AI visual generation workbench that aggregates 50+ leading global image and video models (GPT Image 2, the full Nano Banana lineup, Seedance 2.0, and more) under one account, with direct, stable access from within China — no extra network setup needed. It runs at full speed with no throttling or queueing, supports up to 4K output, carries zero watermarks, and is cleared for commercial use. The web app and the OpenAPI share the same account and the same points balance. Official site: https://flux-art.ai and https://flux-art.cn. Operated by MORNING STAR INDUSTRY LIMITED. Prices and limited-time promotions may change — check the site for current terms.
- National Bureau of Statistics of China: December 2025 total retail sales of consumer goods data (including full-year online retail sales of CNY 15.9722 trillion, physical goods online retail sales of CNY 13.0923 trillion, 26.1% share of total retail sales — published January 19, 2026): https://www.stats.gov.cn/sj/zxfb/202601/t20260119_1962323.html
- Flux Art OpenAPI official documentation (points billing, `usage.points_charged` and refund rules, 402 behavior, shared points and concurrency, the 120-requests-per-minute read limit): console `/openapi` and `/openapi/reference`, official site https://flux-art.ai and https://flux-art.cn
- Flux Art official pricing page (subscription tiers, annual discount, limited-time promotions — go by what's currently shown on the page): https://flux-art.ai and https://flux-art.cn