Most pricing pages tell you how many credits you get. Very few tell you what a credit is worth once you start working. This is the version we wish existed when we were comparing tools: every model on the platform, what one picture costs, and where the money stops buying anything you can see.
The short version
A single image on text to image costs between 2 and 25 credits. That is a twelve-fold spread, and it is not a quality ladder — it is a set of different trade-offs. The cheapest models are fast drafting engines. The most expensive ones spend longer reasoning about your prompt before they draw anything.
On the Lite plan, 1,000 monthly credits buys roughly 500 drafts at 2 credits each, or 40 top-tier renders at 25. Most people land somewhere in between, because most projects are mostly drafts with a handful of finals.
Every model, by what it costs
These are live numbers from the model catalogue, not estimates. Where a model shows a range, the price depends on a setting you choose before you run it — usually resolution or quality.
| Model | Credits | What it is for |
|---|---|---|
| Qwen Z-Image | 2 | Fastest drafting, composition tests |
| Midjourney Diffusion | 2 | Stylised drafting, mood exploration |
| GPT Image 2 | 4 / 8 / 12 | 1K, 2K or 4K; reliable general work |
| Ideogram | 4 / 10 / 15 | Turbo, Balanced or Quality; text inside images |
| Grok | 5 | Quick cinematic stills |
| Wan 2.7 | 5 | Batches — it can return up to four at once |
| GPT Image 1.5 | 5 / 25 | Medium or high; high is the priciest single render here |
| Nano Banana, Nano Banana 2 Lite | 6 | Everyday photographic work |
| Qwen | 6 | Tunable output — steps, guidance, seed |
| GPT Image 2.5 Flare / Sunburst | 6 / 10 / 16 | 1K, 2K, 4K; Flare is quicker, Sunburst is more careful |
| Flux Kontext | 6 / 15 | Pro or Max; editing that respects the original |
| Seedream 5 Lite | 7 | Cinematic stills on a budget |
| Seedream 4.5 | 8 | Product and lifestyle photography |
| Nano Banana Pro, Nano Banana 2 | 8 / 12 / 18-20 | 1K, 2K, 4K; the high-resolution workhorses |
| Seedream 5 Pro | 11 / 21 | Basic or high quality |
| Wan 2.7 Pro | 15 | 4K batches |
Resolution is the expensive dial, not the model name
Look again at the middle of that table. Nano Banana Pro at 1K costs 8 credits. The same model at 4K costs 20. You did not switch to a better engine; you asked the same engine to paint more pixels.
That matters because most images never need 4K. A blog header, a social post, a marketplace thumbnail — all of them get resized down before anyone sees them. If the final placement is 1,200 pixels wide, a 4K render is 10 credits spent on detail that a browser throws away.
Our rule of thumb: draft at 1K, finish at the size the image will actually be used at, and only go to 4K when something will be printed or cropped into hard.
When 4K earns its keep
Print, obviously. Also any image you expect to crop aggressively later — a wide shot you will cut a square and a vertical out of. And product photography where a customer can zoom. Outside those, the extra pixels are a habit, not a requirement. If you need a smaller file from a big render afterwards, resize and compress do it in the browser without spending anything.
Speed costs money, and sometimes saves it
Cheap models are usually fast models. In our own timings, the 2-credit engines come back in roughly 8 to 15 seconds. Mid-range models sit around 15 to 30. The heaviest runs — 4K, or the models that think before they draw — can take a minute or more.
| Tier | Typical wait | Sensible use |
|---|---|---|
| 2-5 credits | 8-20 seconds | Ten variations of a composition, fast |
| 6-12 credits | 15-40 seconds | The version you will actually publish |
| 15-25 credits | 30-90 seconds | Hero images, detailed edits, print |
The saving is not in the credits, it is in the iterations. Five cheap drafts to find the right composition, then one expensive final, costs less than five expensive attempts at guessing.
What a plan buys in pictures
On pricing, the Lite plan's 1,000 monthly credits translate to about 166 everyday renders at 6 credits, or 125 GPT Image 2.5 images at 1K. Pro's 5,000 credits is roughly 830 everyday renders. Max starts at 10,000 and the slider runs to 140,000 on Ultra.
Two details are worth knowing before you choose. Subscription credits refresh each month and do not roll over, so a plan should match your normal month rather than your busiest one. Credit packs work the other way: they are bought once, stay valid for a year, and stack on top of a subscription — which makes them the better fit for seasonal work.
The costs nobody puts on a pricing page
Rewrites. The single biggest drain on a credit balance is not the model you picked, it is running the same flawed prompt six times with slightly different adjectives. A prompt that names the subject, the lighting and the framing usually lands in two attempts. One that says "make it better" never converges.
Wrong starting point. If you already have an image — a product photo, a sketch, a previous render — starting from image to image instead of describing it from scratch is both cheaper and closer to what you wanted. We wrote about choosing between the two separately.
Cleanup you did not need to pay for. Removing a background, dropping an object, changing a backdrop colour — those are dedicated tools, not prompts. Asking a generator to redraw a whole image because of one distracting element costs a full render every time.
A worked example
Say you need one hero image for a landing page and six supporting shots for a product listing.
Drafting the hero: four Qwen Z-Image runs at 2 credits to test framing — 8 credits. Final hero on Nano Banana Pro at 2K — 12 credits. Six supporting shots on Seedream 4.5 at 8 credits each — 48. One background cleanup on the main shot — 4. Total: 72 credits, comfortably inside a Lite month, with the hero rendered at a size you can actually crop later.
The naive version of the same job — eight attempts at 4K on the most expensive model — costs about 200 credits and produces a worse hero, because the framing was never tested cheaply first.
Start cheap, finish expensive
That is the whole method. Explore with the 2-credit engines, decide, then spend once on the model and resolution that matches where the image will live. Every new account gets starter credits, which is enough to run the drafting half of that loop before deciding on a plan.
If you would rather see the models side by side before committing, the creation workspace keeps the same prompt box for all of them, so switching engines is a dropdown rather than a new account. For the frame itself, our guide to aspect ratio and resolution covers the settings that decide how much you spend.
What the free credits actually get you
A new account starts with a small credit grant — enough for a few renders, not enough to finish a project. Spent well, that is three drafts on a 2-credit model and one everyday render at 6, which is exactly the loop this article recommends: explore cheap, commit once.
Spent badly, it is a single 4K render on an expensive model that comes back framed wrong. The starter grant is a test of the method more than a test of the platform.
Plans and packs behave differently
| Subscription credits | Credit packs | |
|---|---|---|
| Arrive | Every month | Once, when you buy |
| Expire | At the next monthly refresh | After a year |
| Roll over | No | Not applicable — they sit until used |
| Stack with a plan | — | Yes |
| Best for | Steady monthly output | Bursts, seasonal work, campaigns |
The practical consequence: size a subscription to a normal month, not your busiest one, and cover the spikes with a pack. Paying for a bigger plan to survive one launch month means buying credits in every quiet month too, and those do not carry forward.
Three ways people overspend
Rendering at the wrong tier by default. Leaving everything at 4K because it sounds safer is the single most expensive habit available. Most images are displayed at a fraction of that.
Regenerating instead of editing. A render that is right except for one distraction should go to a cleanup tool, not back to the generator. Regenerating rerolls everything, including the parts that worked.
Exploring on expensive models. The 2-credit engines exist precisely so that being wrong is cheap. Using a 20-credit model to decide whether you want a wide or a tall frame is paying a specialist to answer a question any model could.