Privacy guide

Privacy in cloud AI image generation

A useful privacy claim should tell you where your content goes, who can connect it to you, and when it is deleted. “Private” by itself is not enough.

AI image and video generators can feel like local creative tools, but most advanced models run in the cloud. A generation request may include a written prompt, reference photos, video or audio, model settings, account metadata, and network information. The service's architecture determines which parties receive each part.

That does not make every cloud AI tool equally private. There is a meaningful difference between sending the minimum content required for one job and building a public, account-linked archive around every creation.

Four questions reveal most of the story

Is the work public?

Some products publish creations to a community gallery by default. Others keep results visible only to the user.

Is history stored in the cloud?

An account-synced library is convenient, but it also creates a durable profile of prompts and media.

Who sees your identity?

A direct provider sees your connection and account. A proxy can separate the end user's direct identity from the model request.

What is the retention window?

Generated files and task logs often have different deletion schedules. Look for both, not just one number.

How VeilPix approaches those questions

VeilPix uses cloud models, so prompts and necessary reference media leave your device for generation. Its privacy design focuses on limiting the surrounding exposure:

The retention detail matters: Kie.ai currently says generated media files are stored for 14 days and then deleted automatically. Its documentation separately says task logs containing text and metadata are stored for two months. A video saved in your local VeilPix gallery can therefore stop playing when the provider file expires, unless you downloaded it.

Read Kie.ai's current task-log and file-retention documentation.

Identity privacy and content privacy are different

Proxying is useful because the model gateway does not receive your direct VeilPix account identity or browser IP in the generation payload. It is not a disguise for the content itself. A face, address, document, location, voice, or unusually specific prompt may still identify a person.

VeilPix's own infrastructure and necessary service providers also receive some information. Clerk handles authentication, Supabase stores account and usage records, Stripe processes payments, and Google Analytics receives configured site analytics. This is why an honest policy should say “data minimization and separation,” not “zero knowledge.”

Published defaults vary across major products

Midjourney says creations are open by default and may appear on Explore, with Stealth Mode limited to higher-priced plans. OpenAI says content from individual ChatGPT services may be used to improve models unless the user opts out, while business and API offerings follow different defaults. Google's Gemini API terms distinguish unpaid services, where submitted content may be used for improvement and human review, from paid services.

VeilPix's distinction is narrower: it does not publish a gallery, does not keep an account-synced creative history, does not train a VeilPix model, and does not attach your direct VeilPix identity to the Kie.ai request. Those strengths do not cancel Kie.ai's or an underlying model provider's own processing and retention rules.

Good habits still help

  1. Remove EXIF location data before uploading sensitive photographs.
  2. Avoid including identification documents, private addresses, or confidential client material.
  3. Download provider-hosted video results you intend to keep.
  4. Delete local gallery items or browser site data when you no longer need them.
  5. Recheck provider policies before especially sensitive or regulated work.

Read the complete VeilPix data flow

The privacy page lists each data category, service provider, retention statement, and user control in one place.

Open VeilPix Privacy & Data Practices