AI & safety
The models behind Z-Image.win, how we check content, and what to consider before sharing an image.
Last updated
An independent creative service
Z-Image.win provides an interface, generation workflow, credits and content checks built around third-party AI services. We are independently operated and are not the official Z-Image model website or a product endorsed by Alibaba, the model developers or our API providers. Third-party names and trademarks belong to their respective owners.
Which models and services we use
Image generation currently uses Z-Image Turbo through EvoLink. Z-Image Turbo is the image model; EvoLink provides the API access used by our service. We do not claim to have developed or own the underlying model.
Prompt and generated-image safety checks use Meta’s Llama Guard 4 12B through OpenRouter, with DeepInfra serving the model. The Image to Prompt tool is configured to use Gemini 3.8 Flash through Kie to describe the image you submit. These are separate tasks from creating a new image.
Availability can change. We update this page when the models offered to customers change; submitted generation tasks retain their recorded provider, settings and credit cost.
Review an image before using it
AI can invent details, render text incorrectly, distort faces or hands, and produce results that differ from your prompt. An image is a synthetic depiction, not proof that a person appeared somewhere, did something or endorsed a product. Do not rely on an output as factual evidence or professional advice.
You may use generated images, including commercially, subject to applicable law, third-party rights and the terms of the relevant model and service providers. We do not guarantee exclusive rights, copyright protection or freedom from third-party claims, and we do not conduct a rights-clearance review of each output. Obtain any permissions needed for third-party material, trademarks or real-person likenesses. Review the rules of the platform where you publish, and disclose AI use when required or when omission would mislead people.
Content rules
These rules apply to prompts, uploads, generated images and their use. Submit only material you own or have permission to process, and obtain permission for the way you intend to depict a real person.
- No sexual exploitation, pornography, explicit sexual nudity or non-consensual intimate imagery.
- No sexualized or exploitative depictions of minors, including fictional or AI-generated depictions.
- No graphic violence, cruelty, hateful abuse, promotion of terrorism or content encouraging self-harm.
- No deceptive impersonation, false endorsements, fraud, harassment or privacy violations.
- No infringement of copyright, trademarks or likeness rights, or attempts to bypass safety, access or payment controls.
How content checks work
The image generator checks your prompt before submitting a generation. Generated images are held in private storage and checked before delivery. A blocked or unresolved result is not shown through the generation and download interfaces. If the checking service is unavailable, the request can be held or fail instead of being automatically approved.
Automated checks can miss harmful content or reject acceptable material. They do not verify identity, age, consent, copyright or factual accuracy, and we do not promise that every violation will be detected. These input and output checks describe the image-generation workflow; they are not a claim that every upload or Image to Prompt request receives the same screening.
A rejected prompt does not start a generation or consume generation credits. A completed upstream generation can still consume credits when its output is withheld under our content rules. The Credits & refunds page explains recovery and checking-service failures.
Report a problem or appeal a decision
Email support with the subject “Safety report” or “Safety appeal”. Include a task ID or page link, a short description, the rule or right involved, and a way to reply. If you represent a rights holder, explain your connection to the material.
Do not attach illegal or intimate material, identity documents, passwords or private download links. We may ask for appropriate evidence to investigate. Depending on the facts, we may restrict access, remove content or reconsider a decision. Reporting content does not automatically trigger a refund.
Contact support
For account, content, privacy or payment questions, email:
support@z-image.win