Horizon Lens — How to judge an image or video before sharing it
First identify the claim
A striking video arrives in a group chat. Its caption says a familiar local bridge has collapsed, and someone asks you to warn everyone nearby. Before deciding whether the pixels look real, separate the claim into parts: which bridge, what happened, and when? This hypothetical example runs through a practical editorial guide to checking images and video before sharing them. The aim is not to turn you into a forensic investigator. It is to stop an impressive picture doing more evidential work than it can support.
An image might show a damaged structure without establishing the location, date or cause named in its caption. Start by writing the claim in one sentence and identifying its source. Was the person who sent it present, or are they forwarding someone else’s post? Ask for the original link rather than another screenshot. Keep the content and its accompanying claim together while checking, so you do not accidentally verify one thing and repeat something else.
Understand what ordinary editing can do
A September 2026 Gradio tutorial describes a single workflow containing image generation, editing, background removal and image-to-video generation. One example takes an uploaded picture and a written instruction to produce an edited image; another animates a still. These are documented creative capabilities. They are not evidence that any particular picture in your inbox was generated, but they give a reason to investigate the source rather than relying only on visual plausibility.
For the bridge clip, resist making the entire decision about a strange shadow or an unusually smooth surface. You can note an inconsistency as a question without treating it as a verdict. Equally, a convincing-looking scene does not answer the questions in the caption. Keep two separate notes: what the file appears to show, and what you have independently established about the event. A gap between those notes is a reason to slow down.
Consider benign explanations as well as deceptive ones. The clip might be a creative demonstration, a reconstruction or a piece of entertainment recaptioned by someone else. Your sharing decision depends on the claim being made now, even when the creator’s original purpose was harmless. Ask where the misleading interpretation entered the chain instead of assuming the first person who made the image intended to deceive.
Trace the version you are looking at
Open the original post where possible. Read the surrounding text, account description and any linked explanation. Look for an earlier or fuller version of the material, including frames before and after a short clip. An image search can be a useful lead to another appearance of the same scene; treat what it finds as a place to investigate, rather than an automatic authenticity certificate. The practical goal is a traceable account of where your copy came from.
Metadata can provide context too. The Gradio tutorial describes writing generation parameters into a PNG image and reading them back, including its prompt and model. That is an example of information a particular workflow records. It does not mean every generated image carries those fields, or that the absence of such fields establishes a photograph’s authenticity. Avoid turning an optional trace into a universal test.
In our hypothetical bridge case, an earlier post might identify the scene as a visual-effects exercise. If it does, preserve the link and the wording that changes your understanding. If you find nothing, record the origin as unresolved. Failure to locate an earlier copy is not positive evidence that the clip documents a new real event.
Verify the event beyond the file
Now check the bridge claim through information that does not simply repeat the same upload. Look for a statement from the relevant operator or local authority and reporting that explains what was independently established. Read dates carefully. Several accounts sharing one clip are several copies of a claim; do not count them as several independent confirmations unless their reporting gives you a reason to do so.
Match the corroboration to the question. Confirmation that a bridge is closed does not necessarily establish that it collapsed. A weather warning does not verify the location of a flood video. If only part of the caption is supported, keep the uncertainty attached to the rest. It is often possible to share a useful official update without forwarding the dramatic footage at all. That preserves the practical information without amplifying an unresolved visual claim.
If you consult an automated assessment, use it as one input, not permission to stop checking. Do not present a confident detector label as proof without understanding what the tool actually evaluated. For material involving identifiable people, allegations or urgent instructions, an unresolved origin should weigh heavily in your decision to withhold the file. A quick warning can still be accurate if it points people to a verified source.
Choose what you can honestly share
Action
Before forwarding an image or clip, answer four questions: what is the claim, where did this version originate, what independently supports it, and what remains unknown? For our bridge example, you might conclude that there is a verified closure but no verified collapse. Share the closure notice with its source. Leave the speculative caption and unexplained video out. That is a useful contribution even if you never determine how the clip was made.
Action
If you already shared an unsupported claim, return to the same conversation with a clear correction and the better source. Explain what changed rather than quietly deleting your certainty. Keep a small record of the links you used for any consequential decision. The habit worth developing is not spotting every artificial image at a glance. It is making the strength of your statement match the strength of the evidence you actually have.