A Candy AI alternative for commercial use and programmatic access

Our reading of Candy AI is from candy.ai itself on 11 September 2026: its Terms of Service (refunds included, in section 10.4), community guidelines and homepage, all documents Candy AI can revise. Candy AI is a live, maintained product: its Terms of Service were last revised on 17 August 2026. The question here is narrower than whether it works. It is whether you are permitted to do what you planned with the output.

Read on 11 September 2026, Candy AI's Terms of Service state that "Our Services are intended and made available for personal, non-commercial use only", and expressly prohibit using the platform, the account, or any content generated on it "for commercial purposes of any kind". A separate clause prohibits using generated content "in connection with off-platform artificial intelligence or machine learning technologies, whether or not for commercial purposes". The Terms also grant Candy AI a "perpetual", "sub-licensable", worldwide licence over user content.

That is a clause rather than a filter, and no setting turns it off. If your plan was a client deliverable, a product asset, a monetised upload or a set to train on, the document answers before anything else does. Our terms take the other position in one sentence: rights in the content you generate belong to you to the extent applicable law permits, and Vellria claims no right over it beyond providing the service.

On 11 September 2026 there was no publicly documented Candy AI API to build against: docs.candy.ai and developers.candy.ai did not resolve, and /api, /api/v1 and /developers redirected to the homepage. Their community guidelines separately prohibit accessing the service or creating accounts "using methods other than publicly supported interfaces", which closes the usual workaround as a matter of policy rather than of engineering. A model page such as Seedream 5 Pro or Qwen 3 lists what that model accepts and costs, and image work in the browser starts at the NSFW AI image generator page.

Vellria's site and API are two doors into one product. A key created in the console, sent in a Bearer header, reaches every model through a request that names the model, gives the prompt and sets that model's options. Runs are asynchronous: you receive a generation id and check it until the file is ready or the run has failed. Nothing in our terms confines the output to personal use or keeps it out of other AI systems, and that is the part a commercial pipeline leans on.

In a pipeline, pull prices and accepted sizes from GET /v1/models instead of pinning them in code. The model pages and the pricing table are drawn from that same data, so a change on the server reaches your integration and the site together. The API documentation has every request shape.

Candy AI's Terms describe a subscription billed monthly, quarterly or annually, alongside non-subscription token purchases. The subscription carries a monthly amount of tokens disclosed to you during checkout, which pay for extended features such as image generation; the Terms state that tokens "must be used within their respective billing periods", and that on cancelling a subscription, remaining tokens "will expire and become unusable". Its Terms (section 10.4), read on 11 September 2026, allow a refund request within 24 hours of payment, deny it once more than 20 tokens have been used, and state "We can only refund subscriptions that were purchased via card."

Here nothing is billed on a cycle. You buy credits ahead of use, by card or with a payment method local to your country, and they wait until you start a run: our terms put no expiry on them, so ending a quiet month forfeits nothing. A run the provider fails returns the credits taken at submission on its own, with nothing to file, and a refund of unused credits can be requested through the support page in the console, each request assessed individually under the refund policy.