Telegram media rules: keep useful screenshots, reduce noisy uploads
Set format permissions and caption rules, test albums, and contain dangerous uploads without downloading suspicious files or spreading additional copies.
A hypothetical technical-support group needs screenshots to diagnose problems. It also gets reaction GIFs, long voice messages and unrelated forwarded videos. Turning off every attachment would make the feed quieter while making support harder.
The better starting point is the problem: an unnecessary format, prohibited wording in a caption, or material from an unwanted source. Telegram permissions, Defendy's text filters and human judgment handle different parts of that decision. None of them should be described as a switch that automatically recognizes “useful media.”
Choose the layer before changing a setting
Telegram's group permissions are the first place to look when members should not send a particular kind of content at all. Telegram explains the principle in its group-permissions announcement.
Caption moderation is appropriate when photos or files are welcome but some accompanying text violates the rules. A filter can receive a message, find a prohibited entry and request deletion. That is different from preventing the member from sending the format in the first place.
Human review is still needed for relevance and meaning. A screenshot without a caption may contain an off-topic argument. A file may be allowed as a document without being safe to open. A caption setting does not establish malware scanning, image understanding or speech transcription.
This distinction also makes troubleshooting quicker. If Telegram will not let the member send the file, changing a word dictionary will not fix the permission. If the file appears and then disappears, investigate the moderation action instead.
Design around the task members came to do
In the support-group example, a practical policy could preserve screenshots and diagnostic files while asking members to describe the problem in text. Voice-only explanations could require a short written summary. Recreational GIF exchanges could move elsewhere.
These are hypothetical policy choices, not built-in Defendy classifications. The bot cannot infer that an image is an important bug report merely because you allow photos.
Where attachments support classes or assignments, decide how members can submit work safely during a restriction. The educational-group guide covers that broader workflow.
Before restricting a format, ask:
- Does anyone need it to participate, including people who find typing difficult?
- Is the problem the format itself, or one person's repeated off-topic use?
- Is there a reasonable alternative that members already understand?
- Who can approve an exception, and how will the member reach them?
A clear exception process is better than telling someone to keep trying different file formats until one gets through. If the core support workflow needs documents, a blanket document ban should be a conscious trade-off, not a side effect of “turn off media.”
Set native permissions narrowly
Open the group's information and member-permissions settings in Telegram; the exact menu layout depends on the client. Record the original settings before changing them. Adjust only the formats the group has decided to restrict, then test from a regular member account.
Telegram distinguishes photos, videos, documents, audio files, voice notes and video notes. Test the actual format you mean: a music file is not the same input as a recorded voice note. The current technical distinctions are listed in ChatPermissions.
Also test an image sent as a file. If documents remain allowed, turning off photo sending should not be treated as a universal ban on files containing images. Decide whether that route is useful for uncompressed support evidence or conflicts with the policy you intended.
Administrators and individually restricted members are poor substitutes for an ordinary test member. A successful admin upload does not show what everyone else can do, and a member with an existing mute may still be unable to send after the general rule changes. The member-permissions guide covers that distinction.
What Defendy's media-text check actually checks
In the Mini App, select the group, open the profanity filter and inspect the media-check option. It allows text associated with media to be checked against the selected dictionary. It does not turn pixels or recorded speech into text.
The reviewed input parser collects more than ordinary text and captions: it also extracts supported rich-format text, block captions, button labels and explicit link destinations. Those are additional text inputs, not evidence that the attached file is downloaded and inspected. A supported parser field alone does not establish that a particular filter catches it in the deployed service.
Keep these three tests separate:
- A picture with a prohibited term typed in its caption.
- A picture showing that term inside the image, with no caption.
- A voice note in which someone says the term.
Only the first is the caption-input scenario described by this setting. Do not advertise the other two as covered without separate, verified functionality.
A new profanity module is initially off, but caption checking and message deletion are initially on. The member action can say None while a matching caption still causes the attachment to disappear. None is not the same as “keep the message.”
For a first test, use a separate group, a harmless custom marker and expendable media. Leave member sanctions off. Warnings can interact with existing points and thresholds, and changing the action resets the shared duration control to zero. If you later choose a temporary restriction, explicitly set and verify its duration after selecting the action.
Settings switches save as they change. Reopen the section after an error or interrupted connection to verify what was actually saved. Check the group name as well: a perfectly saved setting in the wrong group will not explain the result you are observing.
An album is a larger deletion risk
Do not promise that a caption violation will remove only the caption. In the reviewed Defendy implementation, when a triggered event includes the media-group item list, deletion is requested for every message in that group.
The practical consequence is that an album can lose several useful images because of one triggering event. It is not safe to assume that the first image or the visible caption is the only thing at risk.
Use a small album of throwaway images to check the actual behavior. Send it once with a permitted caption, then with a test entry. Record which items remain. Repeat the relevant case after editing the caption. A successful single-photo test does not establish album handling.
For valuable submissions, ask members to keep an original until the moderation policy has been validated. That is a workflow precaution, not a claim that the bot backs up attachments.
A compact pre-launch test set
This hypothetical policy allows troubleshooting images and documents while restricting voice notes. The desired outcomes describe that policy; they are not a report of a live product test.
| Test | Desired outcome | If it differs |
|---|---|---|
| Screenshot with no caption | A member can send it and it remains | Check native permissions and other deletion sources |
| Permitted diagnostic document | The file can be sent | Check document permissions and individual restrictions |
| Recorded voice note | The intended Telegram restriction applies | Check the precise format and the sender's role |
| Image with a prohibited test caption | The configured text filter acts | Check the dictionary, media option and saved state |
| Album with a test caption | The effect on every item is understood | Do not roll out until deletion scope is clear |
| The same image sent as a document | The result fits the document policy | Revisit the rule instead of assuming universal image blocking |
Run the text-filter tests from an ordinary member without a Defendy immunity role. Defendy skips Telegram administrators before automatic filtering; some internal roles also grant filter immunity. An admin control test can help explain the difference, but it cannot replace the participant test.
A forwarded image has another decision attached
A forwarded photo can be assessed by its source independently of its caption. Defendy exposes source controls under Anti-spam → Forwards. The media-check switch visible there currently does not change the forward handler's behavior. Switching it off is not a reliable exemption for forwarded photos.
Test an ordinary upload, a forwarded upload and an upload with a prohibited caption separately. The first matching filter can stop later checks, so “the image disappeared” is not enough to identify the rule responsible. Use the forward-filter guide for source-list and hidden-author limitations.
A dangerous upload with a harmless caption
An attachment can be suspicious without containing a prohibited word in its caption. A new member might post a supposed client update and ask everyone to install it, or an upload might expose personal details alongside a threat. Silence from a text filter does not establish that the attachment is safe.
Do not open, run or download a suspicious attachment for initial moderation triage. Do not ask members to inspect it, forward it to another bot or upload it to a third-party scanner on the group's behalf. That can distribute dangerous content or private information further. An unfamiliar extension alone does not prove malware; record why the post is suspicious without inventing a technical diagnosis.
Act on the available context and the report you received:
- Stop distribution. Remove the post if authorized and check accessible repeats, in-group forwards, quoted replies and pinned messages. Check every relevant album item rather than assuming removal of the caption handled the whole set.
- Contain the source. Apply a proportionate restriction if the sender continues. If an administrator is involved and you cannot act, contact the owner; signs of takeover call for the account-and-group response.
- Narrow formats if targeted action is insufficient. An authorized administrator can temporarily change Telegram permissions during a continuing upload attack. Record the old settings, a responsible person and a review time. A photo restriction does not block documents, and exceptions or individual rights need checking.
- Post a brief warning. Ask members not to open or forward the recent suspicious upload without including the file or harmful link. A member who already opened it needs a separate device-security response; deleting the group post does not resolve that risk.
If the on-duty moderator lacks the necessary rights, send the message location to a verified administrator who has them. Defendy's presence does not give that moderator additional Telegram authority. A caption-check switch cannot compensate for missing deletion rights.
Evidence, reporting and reopening useful uploads
Keep the minimum incident details privately: time, message link or identifier, sender, a safe explanation of the concern and the response outcome. Do not copy the harmful contents into a log or screenshot. Do not download suspected illegal material as proof, especially child sexual abuse material. For personal-data exposure, follow the doxxing response.
Use Report according to Telegram's official reporting instructions. Report before removal when immediately available, without delaying urgent containment to find a menu option. Do not forward the attachment into a public conversation to facilitate a report. Record submission separately from any moderation outcome.
The reviewed Defendy handler does not perform OCR, antivirus scanning, NSFW classification or inspection of file contents. Manual removal does not need a prohibited caption to justify it. Telegram handles platform reports through its own procedures; neither reporting nor the bot guarantees removal of downloaded or forwarded copies.
Restore formats after checking that the source is contained, known accessible copies were addressed and a moderator is ready for repeats. Test ordinary sending with a harmless image or document instead of uploading the dangerous original again. If restrictions remain, explain the temporary route for useful submissions and set the next review.
Restore the workflow after a mistake
If members can no longer submit ordinary support evidence, restore the relevant Telegram permission and retest. If submissions arrive and then vanish, pause the mistaken filter and examine the moderation log, where the relevant events were enabled beforehand.
Fix previously applied consequences separately. Disabling a filter does not undo a warning, a mute or a separately configured reputation penalty. Nor does it restore the original deleted album. After fixing the rule, ask the member to resend the useful material rather than repeatedly testing against their only copy.
The final acceptance criterion is simple: an ordinary member can complete the group's core task. If routine screenshots require administrator intervention or repeated format changes, the policy is getting in the way of the work it was meant to protect.
Sources and review boundary
Updated on October 7, 2026 after checking Telegram's permission reference, its official reporting route and Defendy's text-input parser. The Mini App, forward-filter and media-group deletion explanations are retained from the October 5 review. The profanity reference lists the related controls.
The support group and test cases are hypothetical. No live Telegram group, specific client interface or deployed service version was tested for this article.
