Secret Label Cover · öffentlicher Open Skill
Secret Label Cover
Secret Label Cover ist ein öffentlicher Open Skill auf Digen. Öffne ihn, um zu sehen, was dieser Workflow erstellt.
Product
0 0 1
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4284865/2026-10-20/vrvwcpp8.jpg
# Secret Label Cover Edit one existing image so that the sensitive part of it is replaced by flat rounded "Secret" label blocks, in a single unified visual treatment. This is an image-editing job, not a free-form illustration: the identity, pose, framing, background, and lighting of the source image must survive, and only the sensitive region is turned into labels. ## When to use - The user uploads a photo, screenshot, or document and asks to hide, mask, anonymize, censor, or "make it public-safe". - The user asks for "Secret label cover", label blocks, sticker-style redaction, or a before/after safe version of an image. - The user wants to change something in a result this skill already produced: other label text, fewer or more labels, different label colors, a tighter or looser covered area, a different aspect ratio. - The user provides a photo plus a short text instruction (custom label word, how many labels, which region to cover). Do not use this skill when the user only wants a plain blur or mosaic with no labels, when the user wants a brand-new illustration unrelated to an input image, or when the request is to cover faces for the purpose of deceiving people about who is in the picture. ## Inputs and outputs - Source image (required): the picture to edit. Use the uploaded asset as the generation input. Never invent a different person, product, or scene. - Text (optional): either a custom label word (for example "Private", "Confidential", a translated word, or a short code) or a short instruction about how much to cover and where. Text overrides the default word "Secret" only inside the labels; it is never used as an added caption elsewhere in the image. - Output: one edited image, generated with `generate_image`, saved to the canvas. ## Steps 1. **Restate the job in one line.** Confirm the source image, the region that must be covered, and the label text (default `Secret`). If no image was supplied, ask for it instead of generating from scratch. 2. **Read the two visual references before writing any prompt.** Call `read_skill_file(name="secret-label-cover", path="references/style-and-prompt.md")` and `read_skill_file(name="secret-label-cover", path="references/label-design.md")`. Read them in that order and keep them open for the rest of this turn. Do not write the generation prompt from memory. 3. **Inspect the source image (when needed).** If the sensitive region is not obvious from the user's description, call `recognize_image(image_url=<source asset>, prompt="Describe the subject, pose, hands, framing, lighting, background, and list every region that contains private data such as printed text, ID photo, number, address, QR code, barcode, or a screen.")`. Save the answers you need: subject, pose, camera height, background, light direction, and the exact list of private regions. 4. **Read the operating details.** Call `read_skill_file(name="secret-label-cover", path="references/step-details.md")` and follow its checklist for region selection, prompt assembly order, label count, and retry rules. This file also tells you what to do when the user supplies custom label text. 5. **Build the English prompt.** Using the template in `references/style-and-prompt.md`, fill every placeholder with what you actually saw in this image: the subject description, the covered surface, the label layout from `references/label-design.md`, the preserved elements, and the prohibited changes. Keep the prompt specific to this picture; a generic template sentence is not an acceptable final prompt. 6. **Generate the edited image with the image-to-image channel.** `generate_image(prompt=<assembled English prompt>, input_urls=[<source image asset id or uri>], model='i2i.hd.lite', orientation=<see below>, name='Secret label cover', entity_type='edited_images')` - `orientation`: keep the source framing — `portrait` for vertical photos and posters (default), `landscape` for wide images, `square` for square images. Only switch at t