Question 1

What is Jingwei?

Accepted Answer

精卫 Jingwei is an image attribution protection tool tailored for individual use, serving illustrators, photographers, designers, concept artists, and other image creators.
Before delivery or publication, creators can embed invisible attribution declarations and optional visible anti-theft layers. The goal is to significantly raise the technical cost of theft and malicious editing, while preserving verifiable attribution signals for rights holders.

Question 2

Glaze and Nightshade already exist. Why use Jingwei as well?

Accepted Answer

Glaze and Nightshade mainly target bulk scraping and downstream AI-training scenarios after images are posted online. They are relatively less focused on one-to-one misuse scenarios where someone directly takes a single image and edits or steals it. They also require local installation and stronger hardware, especially dedicated GPUs, and may take longer per image with more visible perturbation artifacts.
Jingwei focuses on fast pre-delivery and pre-publication protection for individual creators. It is designed to raise the cost when someone already has your file and tries AI editing, fusion, or redraw workflows, while embedding verifiable attribution declarations.
Glaze, Nightshade, and Jingwei address different threat models and can be used together.
Important: none of Glaze, Nightshade, or Jingwei can provide complete and permanent protection. Choose based on your workflow and always keep original files.

Question 3

Does Jingwei store user images or process them with AI?

Accepted Answer

1) No long-term retention. Images uploaded for Protect, Verify, or AI Inspect Assist are processed automatically and deleted immediately after completion. We do not maintain an artwork repository and do not use uploads for training or commercial analytics.
2) No AI processing. Jingwei does not use AI models to process user images. All embedding and protection methods are based on classical image processing and digital signal processing, with no deep learning and no loading of .pth / .onnx / .ckpt model files.

Question 4

Can watermarking permanently and completely prevent AI fusion, training, or misuse?

Accepted Answer

Invisible layers can be weakened or removed by aggressive compression, repeated transcoding, or targeted removal. Visible anti-theft layers can increase restoration and misuse costs, but no one can guarantee all future tools will fail. The JW declaration communicates ownership and usage intent, and its practical force depends on whether platforms and tools choose to recognize and enforce it.
The JW initiative aims to establish a standardized, machine-readable symbol of creator intent, similar to Creative Commons markers, so platforms can incorporate it into review and usage policies. This is a long-term ecosystem effort and does not replace legal protection.

Question 5

Can Jingwei prevent AI editing and image fusion?

Accepted Answer

Jingwei targets cases where others already possess your image and run AI editing, fusion, or partial redraw on it. Through layered perturbation and optional visible anti-theft overlays, it increases the chance of failed edits or poor restoration, significantly raising abuse cost.
Actual results vary with source image characteristics, toolchain, and post-processing, and may change as technology evolves.

Question 6

Which formats are supported, and which features may affect visual quality?

Accepted Answer

Upload supports PNG, JPEG, and WebP. If you enable JW or other invisible frequency-domain layers, PNG export is recommended because lossy JPEG compression can weaken invisible signals.
JW declarations, DWT watermarking, LSB, and other invisible layers are tuned to minimize normal viewing impact, but stronger settings can introduce subtle texture or color shifts. Visible anti-theft layers such as displacement marks, emboss locks, and blur strips intentionally alter appearance; stronger visibility usually means stronger deterrence. Compliance metadata (EXIF, C2PA, etc.) typically does not affect image appearance.
After processing, the page provides quality reference information. If visual fidelity matters, start with medium or light settings, preview, and export once satisfied.

Question 7

Can verification work on screenshots and images re-downloaded from social platforms?

Accepted Answer

Partially. JW declarations and reinforcement layers include redundancy designed for screenshots, social compression, and secondary saves.
A common path is: protect an image, screenshot it, post to social media, redownload it, and then verify. In many cases attribution can still be detected, but outcomes depend on source quality, screenshot method, and platform compression severity, so full recovery cannot be guaranteed every time.
The Verify page reports per-layer results and confidence for JW, DWT, LSB, anchors, and more. For disputes, retain your original, unredistributed export.

Question 8

How can I verify whether an image has Jingwei watermark data?

Accepted Answer

Open the Verify page and upload the target image (no login required). The system checks JW declaration data, DWT, LSB, EXIF/IPTC, C2PA credentials, and related signals, then reports detection status and confidence by layer.
No detection does not prove that processing never happened. Positive detection can serve as supporting evidence that the image was protected by Jingwei or carries related metadata.

Question 9

Can "AI Inspect Assist" determine whether an image is AI-generated?

Accepted Answer

No final verdict is provided. AI Inspect Assist (beta) outputs a practical "ingredient list" showing whether files contain AI-tool metadata, camera EXIF, C2PA credentials, and similar traces for user judgment.
Files downloaded from social platforms or captured via screenshots often lose metadata. "Not found" does not mean "definitely human-created."

Question 10

Can I change the watermark font?

Accepted Answer

Due to font licensing constraints, only open-license fonts are currently supported and custom font switching is not available yet. You can still adjust watermark size, position, and visible-layer styles. If additional eligible fonts are introduced, updates will be announced on the Protect page and in this FAQ.

Question 11

What is the JW declaration? Is it mandatory?

Accepted Answer

JW (Jingwei Protocol) writes creation type (Original / AI-Assisted), usage restrictions (such as no unauthorized AI training or AI editing), and creator name into the image in machine-readable form, with an optional visible badge.
Our goal is for JW, like CC markers, to become a broadly recognized public signal that platforms and tools can use in review and usage governance, reducing infringement from the mechanism layer. This is a long-term direction requiring ecosystem consensus.
It is optional, not mandatory. If you want to preserve invisible layers, PNG export is recommended.

Question 12

What does "Commission Delivery" do?

Accepted Answer

Commission Delivery provides a protected handoff workflow. Creators upload the final image, set a viewing password, and receive an encrypted delivery package. Clients must open the package on the site's viewer page, enter the password, and hold mouse or touch to progressively reveal the image; releasing input scatters it again.
The viewer also overlays a slowly moving watermark showing viewer identity. Combined with hold-to-view behavior, this reduces the chance of leaking a complete high-resolution final via screenshots. Touch is supported on mobile browsers, though complex operations are still easier on desktop.

Question 13

Why is there no open-source release yet? Will there be a mobile app?

Accepted Answer

Jingwei is still in internal/beta evolution and no public open-source release is available yet. After core features stabilize, the plan is to gradually open source components that do not expose core algorithms or encryption internals, enabling community review and collaboration.
A mobile app requires substantial redesign, app-store compliance, and ongoing operational overhead. There is currently no app roadmap; web experience remains the top priority.

Frequently Asked Questions