AI output and failure clause

What our AI pipeline is and is not. What happens when it refuses, fails, or produces a bad result. What we do and do not do with your data — stated honestly, including where we cannot give an absolute guarantee.

Version: 1.1Effective: 7 November 2026Last reviewed: 7 November 2026

1. What an AI output on ZZEESS actually is

Every image, tile, motif, composite, or recommendation produced by the platform is a candidate — a machine-generated proposal from your brief. It is not:

  • a photograph of a real object that exists today;
  • a guarantee that a workshop can produce exactly that garment;
  • a legally binding attestation of a craft's authenticity;
  • a priced quote for making the garment;
  • a representation that any person depicted exists, is available, or has consented to the use.

A candidate is a starting point for your work — it is not the work. A person always chooses what to keep; the AI does not make a decision about you on its own (see the privacy policy, automated decisions).

2. Why we never name the model to you

We integrate multiple AI vendors at different points in the pipeline, and we change which one serves which function as the technology moves. We do not disclose which specific model produced a given render, and a model's brand name does not travel on any customer-facing surface. This is a deliberate product decision. It does not reduce our responsibility to you: your contract is with us, we remain accountable for every vendor we use, and the privacy policy discloses the categories of vendor and the regions — including outside India — that your content can reach.

3. What we do and do not do with your data

We do not:

  • use your prompts, designs, or renders to train or fine-tune any model, ours or a vendor's;
  • share your business's generated work with other tenants, except via explicit grants you make;
  • sell, rent, or syndicate your personal data to anyone for marketing;
  • store Try-It-On source photographs on our systems beyond the single render;
  • contact your clients from information you upload about them, unless you ask us to.

We do:

  • store renders you ask us to keep, under your business account or (for B2C) your personal account, for the retention period in the privacy policy;
  • read a design back with a vision model (the "Observer") to check whether what was rendered matches what was asked for — only on your own data, and visible only to our operators;
  • send prompts and images to third-party AI vendors to produce the output, as the privacy policy describes — and require, through our vendor terms, that they use your content only for that operation and do not train on it.

What we require, and what we cannot promise. We instruct every AI vendor not to train on your content and are putting written data-processing terms in place to bind this. We cannot inspect a vendor's internal systems, so we do not pretend to give an absolute guarantee of their conduct. Instead we disclose who they are by region, keep what we send to the minimum the feature needs, and treat your use of these features as your informed consent to the transfer. If a vendor's terms will not support these commitments, we stop sending it personal data.

4. When the AI refuses

A vendor may decline a specific request on safety, policy, or content grounds — a prohibited subject, a safety-filter trigger, a content category outside its scope. When that happens, the job is marked as refused and you are not charged. We surface the refusal without translating it; the words that come back are the vendor's, because paraphrasing a refusal is where misleading explanations are born.

5. When the AI fails — the failure clause

A failed render is one where the vendor accepted the request, charged us, and did not return a usable result (a timeout, a corrupted output, a vendor outage). In every such case:

  • the job is retried up to three times at our cost;
  • if the retries do not succeed, the wallet charge that corresponds to the failed render is automatically reversed — a credit back to the same bucket it was debited from (promo credits return to promo, paid credits return to paid);
  • we do not charge a service fee on a failed render.

We do not guarantee a specific accuracy or availability for AI output, because it depends on third-party models whose behaviour we do not fully control. We do guarantee the credit mechanism above.

6. When the output is wrong but technically delivered

If a render finished but the result is unusable — wrong garment, wrong colour, incoherent — ask for a re-render through the module or email support@zzeess.com. For paying accounts we do not charge for a credited re-render of a clearly broken result within 7 days of the original.

7. Faces and likenesses (Try-It-On)

A Try-It-On flow uses a photograph you upload to composite a design onto a person. A photograph of an identifiable person is sensitive data, and we process it only on the basis of consent. By uploading, you confirm that the photograph is of you, OR that you have the explicit consent of the person depicted, for this use.

To make the composite, the photograph is sent to our virtual try-on vendor (see the privacy policy, section 6, for who these vendors are and where they are located). On our systems the source photograph is deleted immediately after the render and is never stored, re-used or used to train models. The photograph is sent to the vendor with the vendor's own storage switched off, so it is not retained there after the render — that is something we control, in the request itself, and we do it. We also require that it is not used to train any model; we cannot audit a vendor's internal systems, so we do not promise what we cannot see, but switching off its storage we can. No Try-It-On may ever depict a minor.

8. Admin-shared creatives

Some packages give B2B customers access to Chhapai and Karigar creatives we curate centrally. These are AI-assisted and human-curated by us; copyright in them belongs to us; a licence is granted to the subscribing tenant under their package. Raw files (PSD / TIFF) are a paid add-on.

9. Reporting an AI problem

If an output on the platform appears to depict a real person without consent, violates copyright, or breaches our Acceptable Use policy, write to our Grievance Officer at grievance@zzeess.com. We acknowledge within 24 hours and act within 72 hours or the timeline required by law, whichever is shorter.

Munshee Digiventures Pvt Ltd

B-12 Balkrishna Bldg, Bh. Presidency School, Adajan, Surat 395009, Gujarat, India

CIN: U72900GJ2019PTC109067

Questions about this page? Email support@zzeess.com or see the contact page.

AI output and failure clause — ZZEESS AI