The generate button shows a credit cost because upstream image models are not free to run. Nanomelon converts that cost into a single balance so you are not managing five vendor dashboards. This page explains what you are paying for, what you are not paying for, and how to spend a small free grant without learning the wrong lesson.
If the job fails or the model refuses the prompt as a policy violation, credits are not deducted. If the job succeeds and you hate the picture, credits are deducted. That distinction matters: “I will click five times and pick the least bad” is a paid lottery. “I will click once, change one clause, click again” is an editing session. The FAQ on the homepage states the refund-on-failure rule because people reasonably assume the opposite after using other apps.
New accounts receive a small starter amount. It is enough to learn the five-part prompt and the two modes. It is not enough to brute-force a style by generating a grid of twenty near-duplicates. When the balance is gone, you can buy a package from the pricing modal or stop. Browsing guides and the public gallery does not cost credits.
Each option in the model list is a different upstream engine with its own defaults: how strongly it follows short prompts, how it draws hands, whether it prefers illustration or photography, how aggressive its safety filter is. A prompt written for a painterly model will look over-specified on a photographic model, and a prompt written for a photo model will look empty on a stylized one. The gallery records the model name on each public page so you can see which engine produced the look—not so you can treat every name as a brand we own. We are not those vendors; we call their APIs.
If a model disappears from the list, it was usually retired or repriced upstream. Do not bookmark a name as eternal. If you need reproducibility for a client, save the prompt, the ratio, the mode, and a still of the output, not only the model string.
The form’s ratio control is the crop. 1:1, 3:4, 4:5, 9:16, 16:9 and the rest map to typical social and print shapes. The generator also has a pixel size under the hood (you will see width and height on gallery detail pages, for example 1024×1536). A tall 3:4 is the right choice for a festival character who should fill the frame. A wide 16:9 is the right choice for a street full of lanterns where the architecture is the point. Using the wrong ratio and then cropping in another app throws away the composition the model spent its attention on.
Quantity multiplies cost. Two images is two completed jobs. Use quantity greater than 1 only after the prompt is stable. The count control (how many times to run the pipeline) is the same idea: it is not a quality slider.
The public-visible toggle decides whether a finished image may appear in the public grid and, if we generate a static page, in the gallery URL list. Do not publish pictures you would not want a stranger to open with the prompt in plain text. Do not publish other people’s faces without permission. Private works remain in your account for download and for your own img2img chains.
That plan is slower than mashing the button and more respectful of a starter balance. It is also how you get a gallery page worth reading: a prompt that explains the picture, not a mood fragment. Compare a short portrait prompt with a long festival prompt—the second leaves less for the model to invent, which is why the notes on that page can talk about specific objects.
Commercial use of outputs, for paying users, is described in the Terms and the homepage FAQ. You still cannot infringe someone else’s trademark or likeness just because a model will draw it. If a job is blocked, rewrite toward an original character or a scene you have the right to depict.
Ready to spend a credit on purpose? Open the generator. For interface steps, use the user guide.