Tencent Yuanbao AI Recorder Adds 'Record and Snap' Feature, Photos Auto-Inserted into Transcript

When someone says "everyone look at this PPT" in a meeting, most people look up and remember the image, but after the meeting when they open the recording, all that is left are references like "this line on the right," "the third plan," and "the item marked in red." This scenario is exactly the starting point of QbitAI's hands-on article: in the past, voice recorders rarely seriously touched the camera, not because everyone forgot there were PPTs in meetings, but because photos and audio belonged to two different folders, and in the end a person still had to step in and pair them up.

From this the author puts forward a judgment: installing a lens was never difficult, and phones have long been able to record audio and take photos at the same time. What is truly difficult is letting a photo know which sentence it belongs to, which moment of discussion it belongs to, and what can still be done with it after the meeting. When this relationship is not sorted out, the camera only produces an extra batch of JPGs.

The entry point for Yuanbao AI voice recorder is not complicated: users on the new version open the APP and can see the feature above the dialog box, then tap to record; users on the older unupdated version can enter the voice recorder from the "+" on the home page. After recording starts, real-time transcription quickly begins spitting out text.

In the test, the meeting discussed the content launch plan for a new smart desk lamp. When the speaker said, "the first 4 weeks were basically flat, it started to rise in week 5, and by week 8 the estimated completion rate reached 68%," the author tapped the camera icon in the real-time transcription page and took a photo of the blue curve. When the shutter was pressed, the recording did not stop, and transcription continued; after returning to the page, the speaker's second half sentence had already been recorded.

The PPT photo just taken was directly inserted into the transcription segment from moments ago. After the meeting, tapping into the earliest photo of the blue curve and selecting "locate to transcription position" made the page jump directly back to the sentence "it started to rise in week 5"; tapping the timeline and budget table also returned to the minute in which each appeared. Photos taken on site and photos added after the meeting can be managed at the same time, and up to 10 can be added in a single session.

After about two minutes, the real-time summary began generating, summarizing the background and data just discussed. The author described it by saying that while he was still struggling to keep up with the PPT, it had already quietly started doing the work of a scene recorder beside him.

More critical is asking questions during the meeting. When the three plans A, B, and C on the PPT were discussed to the point of "let's not do the third plan for now" and the author forgot to take a photo, he took advantage of the recording still being in progress and directly asked Yuanbao: as of now, what conclusions has the meeting confirmed, and what questions remain undecided? Yuanbao quickly summarized the settled content and unresolved questions, and after returning to the recording interface the record was completely uninterrupted. As for the plan page he missed photographing, by long-pressing the corresponding transcription and adding the screenshot of page 5 obtained after the meeting, the image also returned beside that sentence.

The author sorts out a line of product evolution: as sound pickup, noise reduction, and storage matured, voice recorders first preserved sound; as speech recognition matured, they began outputting transcriptions; only when large models could summarize and answer questions were speaker recognition, key point extraction, and minutes generation arranged one after another. Each change means that after the underlying capability is upgraded, it takes on one more kind of work.

In his view, this Yuanbao upgrade is clearly not just an extra camera entry point—the camera may never have been the point; the point is that the voice recorder finally has the ability to process the content the camera brings back. It puts images back at the corresponding time point, then connects the time point with the original words, audio, and transcription, so photos are no longer just an image that suddenly appears after the meeting, but native information within the on-site context.

It should be noted that this layer of generalization—"moving from single audio toward voice, text, and image"—is an inference made by the source article based on a single test session, not an official technical explanation.

After the recording ends, content in Yuanbao's meeting minutes can be edited directly, and any inaccurate wording can be corrected on the spot. Work that originally required moving back and forth among recordings, photo albums, and documents becomes one generation plus one proofreading.

After the recording ends, the content can also be processed a second time, for example to generate minutes, to-do lists, knowledge summaries, and so on, and document forms such as Word, Excel, and PDF can be selected as needed. The recording's audio, transcription, and summary can all be exported, and text records can be chosen as DOCX, PDF, TXT, or Markdown.

Ecosystem connection is a point the author especially emphasizes: meeting records can be sent to platforms such as WorkBuddy and Tencent Docs with one click for follow-up operations, with no need to download. He casually had WorkBuddy organize detailed minutes based on the same record, a management summary retaining only conclusions and risks, and a to-do list rearranged by person responsible.

After switching the audio source to phone-internal audio, the content being played on the phone can be recorded throughout without turning on the speaker. The author tested playing an English interview while wearing earphones and it could still be recorded, with transcription in both Chinese and English, leaving the original sound, English transcription, and Chinese translation in the same record, so it can be directly located when reviewing.

Yuanbao also distinguishes speakers based on different voices and marks them separately in the transcription. But the author also clearly mentions that when encountering proper nouns, fast speech, or places prone to ambiguity, the transcription will have certain flaws; fortunately, the original sound, transcription, and time point remain together, so wherever there is uncertainty, one can directly locate it and listen again.

The author says all the above full features are completely free. But this statement comes from the source article itself, and the source did not explain the scope or duration of the free offering, nor whether charges will be introduced later, so it can only be regarded as a source claim and cannot be treated as a confirmed pricing fact.

The sentence at the end of the article—"I have already begun to feel worried about those standalone voice recorder products costing over a thousand yuan"—is likewise a commentary opinion. The author believes that where Yuanbao's voice recorder built into AI opens a gap with traditional recording devices is that AI does not have to wait for the user to finish uploading materials before cleaning things up; it participates from the moment recording happens—during the meeting you can ask about consensus and unresolved questions, after the meeting you can generate different content based on the same draft, and when advancement is needed you can directly transfer the results to WorkBuddy or Tencent Docs.

The reason this route is "light" is that a phone already has a microphone, camera, screen, network, and computing power, making it a natural carrier for multimodal recording. AI is responsible for reorganizing these ready-made capabilities, so users can complete recording, photographing, transcription, translation, summarization, and document output without buying another piece of hardware. But it should be noted that this judgment lacks support from sales or market data, and the source also did not give the feature's specific launch time, version number, or covered regions, nor did it mention whether it has already launched simultaneously on both iOS and Android.