AI Fortune Reading Methods & Privacy: Photos, Prompts and Storage

How palm and face report generation works

The palm and face services retrieve an uploaded image and send image content with written instructions to a configured DMXAPI gateway using the model identifier gemini-3-pro-image. They request a generated report image with annotations. This is third-party image processing, not an analysis that stays entirely in your browser. The configured endpoint and provider behavior can change; this description does not verify every downstream processor.

The palm prompt references major lines, auxiliary lines, mounts, hand shapes, Five Elements ideas and Mai Yi Shen Xiang. The face prompt names Mai Yi Shen Xiang, Xiang Li Heng Zhen and Liu Zhuang Xiang Fa alongside facial regions and Five Elements associations. These are references in the instructions, not proof that quotations were retrieved from authenticated editions or that the model was trained on a curated specialist dataset.

Both prompts ask for ten analysis points, visible features, soft interpretive language and practical reminders. They instruct the generator to preserve the uploaded subject and avoid medical diagnosis. Those are requested behaviors, not guarantees of pixel preservation or factual correctness. Compare the original photo with the output and treat any health, wealth or relationship claims as unvalidated cultural interpretation.

BaZi uses birth inputs and a text-model prompt

The BaZi service sends a text request to the configured LLM endpoint, with DMXAPI configuration used as a fallback. The reviewed model identifier is deepseek-v4-pro. The message includes birth date, time, selected gender and calendar type, plus supplied city, time zone and solar-time offset. The API accepts longitude too, but the reviewed message builder does not directly insert that field into the model message.

Instructions ask the model to produce structured Four Pillars data and a written interpretation, using conventions such as a Li Chun year boundary, solar-term months and two-hour branches. In this service path, those are model instructions rather than proof of a separately verified deterministic calendar calculation. Check uncertain birth times and boundary dates independently. Fluent prose cannot resolve an incorrect input or a mistaken conversion.

• Read the zodiac and Four Pillars guide

What is uploaded and stored

The direct palm and face APIs upload input photos to Cloudinary in account-associated folders. Their services also upload generated report images to Cloudinary result folders. Database generation-task records associate the account with the input image URL, output URL when available, status, timing and credit information. Failed tasks can retain the input URL for retry. A failed generation therefore does not imply that the photo was never stored.

The general generation upload API is a separate path that also stores media in Cloudinary. It supports account uploads and restricted guest image uploads, with guest identifiers and separate folders. Uploading through that path is not the same as receiving a fortune report, and guest upload support does not mean an anonymous paid report is available. The reviewed direct palm, face and BaZi APIs require sign-in.

BaZi tasks store birth details in their prompt or metadata, including date, time, gender, calendar and supplied city; the API also attempts to persist generated report text in task metadata. Streaming the response does not mean nothing is stored. Service logs can include input image URL fragments or birth date, time and city. Logs, task records and image storage are separate data surfaces.

Retention, access and training: what is not established

The reviewed generation paths do not establish an automatic deletion deadline for input photos, result images or tasks. The presence of a Cloudinary deletion helper elsewhere is not evidence that it runs after each reading. This guide therefore cannot promise immediate deletion, no storage or a particular retention period. It also cannot verify backup expiry or provider-side retention from application code.

Authenticated task lookup checks account ownership, but that is different from access protection for the underlying media URL. The upload helper returns Cloudinary delivery URLs and does not explicitly request authenticated-only delivery in these paths. Treat image and report URLs as sensitive, avoid posting them publicly, and do not assume an unshared link is equivalent to a private authenticated file.

The code does not establish a contractual guarantee that every processor excludes your data from training. It would be misleading to promise never used for training based only on these requests. Review the current privacy policy and ask support about the applicable processors, retention, deletion and training terms before uploading if those issues determine your consent. If the answers do not meet your needs, do not submit the data.

• Read the current privacy policy and support contact details

A practical checklist before and after submission

Use this process for your own information and obtain explicit permission before submitting someone else’s photograph or birth details. Consent to show you a photo is not consent to third-party AI processing.

1. Read the current tool requirements and checkout terms before uploading. Signup rewards should not be interpreted as a promise of a complimentary report.
2. Remove unrelated background details and avoid including identity documents, other people or unnecessary personal information. Preserve the palm or portrait features needed for the chosen tool.
3. For BaZi, check date, time, calendar and location context before submitting. Do not invent a birth time to make a report look more precise.
4. Keep the task identifier if you need support, but avoid copying private media URLs into public forums. Ask what deletion can cover across tasks, media, logs and processors rather than assuming one action erases all copies.
5. Review the output for changed features, unreadable labels and unsupported claims. Never use it for diagnosis, financial decisions or judging another person’s character.

Different outcomes and illustrative examples

If generation fails after upload, the storage step may already have happened. If a report succeeds, both an input and a generated output can exist. If you share the report, recipients may see the original subject embedded in it. These are distinct situations; closing a browser tab does not demonstrate deletion in any of them. Questions about a particular task should go through support with minimal necessary information.

Examples in the new guides are labeled illustrative because they are hypothetical teaching examples, not verified customer reports, testimonials or evidence of accuracy. A sample sentence can explain report structure, but it cannot establish what every purchaser will receive. The tools below provide cultural entertainment; review the current interface and actual checkout terms for the offered service.

• View Palm Reading
• View Face Reading
• View BaZi
• Learn how to review an interpretation