Publish & Pack
Publish & Pack is Nephele's artwork output tool. It can overlay visible watermarks on your originals (image or text, centered or tiled), embed an invisible watermark for provenance that the eye can't see, and export to PNG with full image quality and transparency preserved. It supports batch processing, so you can handle an entire folder in one pass.
技巧
Publish & Pack runs as pure local computation — no network, no stamina consumed. Everything is processed on your own machine.
Where to Find It
| Entry point | Action |
|---|---|
| Toolbox | Left navigation bar → Toolbox → Page 1 → Publish & Pack |
| Agent chat | Just tell Nephele "pack this image for me" or "add a watermark to this image" |
| Drag & drop | Drag an image into the Nephele window, and the Agent can call the packing tool automatically |
How to Use
1. Drop in Images
You can drop in multiple images at once, or click to select files.
Supported formats: PNG / JPG / WebP / PSD / PSB
提示
PSD files are automatically flattened across all visible layers before processing — you don't need to export them manually.
2. Configure the Watermark (optional)
Publish & Pack offers three watermark methods, which can be used on their own or stacked together:
| Watermark type | Purpose | Best for |
|---|---|---|
| No watermark | Output as-is | Final deliverables handed to a client |
| Image watermark | Upload a PNG to overlay as a watermark | Brand logo, personal signature |
| Text watermark | Type text to auto-generate a watermark | Quickly stamping copyright info |
Image Watermark Modes
- Centered — the watermark sits in the exact center of the frame, ideal for a single large logo
- Tiled — the watermark repeats like tiles across the whole image, which is more effective against theft
Watermark Style Adjustments
- Opacity — 0~100%, controls how prominent the watermark is
- Scale — controls the watermark's size relative to the frame
- Rotation — applies a tilt angle to the watermark
- Spacing / offset — in tiled mode, controls how densely the watermarks are arranged
3. Configure the Invisible Watermark (optional)
Beyond the visible watermark, Publish & Pack can also embed an invisible watermark (blind watermark) — writing copyright information directly into the image's frequency domain, completely invisible to the eye, yet extractable with a tool.
How It Works (white-box, fully disclosed)
Nephele uses a blind-watermark algorithm based on DWT-DCT-SVD (Discrete Wavelet + Discrete Cosine + Singular Value Decomposition), built on the open-source library blind_watermark:
- Embedding domain: frequency-domain coefficients after wavelet decomposition, which don't affect visual appearance
- Operating mode: file-level I/O + bit-mode embedding, bypassing known compatibility issues with numpy arrays
- Payload capacity: a fixed 256 bits (32 bytes), roughly 10 UTF-8 Chinese characters or 32 ASCII characters
- Scrambling parameters: a two-parameter system (image scrambling + watermark scrambling) used as a pseudo-random seed to shuffle the embedding positions — this is an obfuscation measure, not cryptographic encryption; it raises the bar for extraction but does not resist reverse engineering
- Alpha channel: PNG/RGBA input has its alpha channel restored intact after embedding
- Embedding verification: immediately after each image is written, a round-trip extraction check runs to confirm the watermark is recoverable; if verification fails it is logged, but the image is still output
Robustness Notes
The invisible watermark has some resistance to the following operations:
- Moderate JPEG compression, scaling, cropping
The following operations may corrupt or destroy the watermark:
- Heavy compression (quality below 60%), major color grading, rotation, or overwriting/cropping more than 50% of the original image's area
注意
The invisible watermark is not an absolutely secure cryptographic protection — it is an aid to provenance. If an image is heavily compressed, cropped, color-graded, or re-encoded, the watermark may be lost. We recommend using it together with a visible watermark and digital timestamping, rather than relying on it alone as the sole copyright safeguard.
4. Choose the Output Directory
By default, output goes to the folder containing the original image, but you can specify another location.
5. Click "Start Packing"
The software processes each image one by one, with progress shown live in the list.
About the Output Files
Each original produces one processed PNG, preserving the original image quality and transparency, with the visible watermark overlaid / invisible watermark embedded according to your settings.
| Case | Output filename |
|---|---|
| Visible watermark configured | {original_name}_wm.png |
| No visible watermark configured | {original_name}_pack.png |
By default, output files go to the original image's directory (or the output directory you specify). The invisible watermark (if enabled) is written directly into this PNG; no separate file is generated.
Extracting the Invisible Watermark
If you receive an image you suspect has been stolen, you can use Publish & Pack's "Extract Watermark" feature to verify it:
- Find the "Extract Watermark" button in the Publish & Pack interface
- Select the image you want to check
- If the image contains a Nephele-format invisible watermark, the copyright info will be displayed
提示
Extract Watermark can only recognize invisible watermarks embedded by Nephele. If an image has been heavily modified (heavy compression, cropping, color grading), the watermark may already be corrupted and unrecoverable.
Using It in Agent Chat
In Agent chat, you can invoke Publish & Pack like this:
"Pack this image for me and add my signature watermark"
"Pack this PSD into a PNG with a signature watermark"
"Add a tiled watermark at 30% opacity to this whole batch"
The Agent will set the corresponding parameters and run the packing automatically.
FAQ
Q: Can I pack a PSD file directly?
Yes. Nephele automatically flattens all visible layers of the PSD before processing, so you don't need to manually export it to PNG.
Q: Can the watermark be easily removed?
A centered or tiled visible watermark can be partially removed if it's cropped or painted over, but a tiled watermark covers a large area, making it costly to remove. The invisible watermark is more discreet, but it can also fail under heavy compression or color grading. We recommend using both together.
Q: How many images can I process in a batch?
There's no hard limit, but we recommend no more than 50 at a time to avoid excessive memory usage. Each image takes about 1~3 seconds to process.
Q: Will the output image's color space change?
Output is PNG, preserving the original color mode and transparency. If your original uses a special color space such as CMYK, we recommend converting it to RGB before packing.
Q: How is it that only Nephele can extract the invisible watermark?
Nephele uses two-parameter scrambling: two fixed parameters each serve as a pseudo-random seed to shuffle where the watermark is embedded in the frequency domain. Extraction restores it using the same parameters, so ordinary tools can't read it. But this is an obfuscation measure, not cryptographic encryption — blind_watermark is an open-source library and the parameters are fixed values, so anyone who knows the algorithm and is sufficiently motivated could still extract it through reverse engineering. The real provenance anchor is the RFC 3161 timestamp signature paired with Digital Timestamping, not this watermark layer itself.
Publish & Pack serves only as an artwork output tool and does not constitute legal copyright protection. For legal verification of ownership, please use Nephele's Digital Timestamping feature.