Upload only the visual content needed for the analysis. Crop or redact private messages, names, account information and unrelated browser interface before submitting a screenshot or screen recording. Do this in your own preparation step; do not assume the analyzer automatically removes sensitive information. A cleaner reference also produces a more focused visual brief.
Decide which region answers the question
If you want the lighting and pose of a portrait, a browser sidebar, comment thread and account avatar usually add nothing. If you want a poster's layout, preserve the poster and remove surrounding messages. For video, check the beginning and end as well as the attractive middle section; notifications or account screens can appear between useful frames.
Use content you own or are authorized to analyze. Do not treat a publicly viewable image as permission to reproduce an identity, watermark or brand asset. When the purpose is to learn a visual treatment, focus on general composition and lighting rather than reconstructing someone else's exact publication.
Redact before upload rather than after the answer
Cover private content with an opaque area or crop it away in a local editor. Blur can leave text partially readable, and a temporary overlay in an editing interface may not be baked into the exported file. Open the actual prepared file to verify what is visible.
Analyze only the prepared visual reference. Describe the visible subject, composition, color relationships and lighting. Ignore the solid redacted areas and do not infer text or personal information behind them. Treat interface elements outside the artwork as irrelevant. Separate observed visual facts from suggestions for a new original image.
That instruction helps define the analysis scope, but it is not a privacy control. Data minimization must happen before transmission. A prompt asking the model to ignore a secret does not undo sending it. Follow the service's current privacy policy when deciding whether any reference should be uploaded at all.
Review a recording frame by frame where necessary
A screen recording can include bookmarks, file paths, download notifications or messages that are absent from a thumbnail. Trim to the needed segment using an appropriate local editor, then watch the exported segment. If audio is included, consider whether it contains names, addresses or private conversation unrelated to the visual task.
- Identify the artwork or scene you actually need analyzed.
- Remove surrounding account and communication content.
- Check for notifications across the selected video segment.
- Export the prepared file rather than uploading a working project.
- Open or replay that file to verify the redactions.
- Review the returned prompt for unnecessary copied names or text.
Keep your unredacted original separately if you need it for personal records. The analysis copy can be a different file with a descriptive name such as prepared-poster-reference. Avoid putting personal identifiers in the filename merely to organize the task. Only submit the amount of material required to answer the question.
Keep visual description and confidential context apart
Runway's image-to-video guide distinguishes the image's visual setup from motion instructions. That separation suggests a useful editorial discipline: provide relevant visual evidence, then specify the intended action separately. It does not establish any automatic redaction feature in Imagild or another service.
Will a prompt saying do not read this text protect a private screenshot?
No. Remove the text before uploading. A model instruction cannot reverse a transmission or guarantee that hidden information is never processed.
Can I analyze only a cropped detail?
Yes, when it still shows the relationships you need. Use Reverse Prompt with a prepared authorized image or video and clearly state when the crop prevents broader conclusions about the original scene.