Insta360 X6 turns filming into supervising a camera agent

Insta360 X6 makes the camera's job less like taking a shot and more like maintaining a usable record. According to geekpark, the new panoramic camera uses a 1/1.1-inch square sensor better suited to panoramic imaging. Its AI Director reviews the previous natural day's footage every night, then edits it before pushing a finished video to the user's phone. It does this without manual upload or manual clip selection.

That changes the operator's work.

The important part is not only that capture gets wider or that editing gets automated. ifanr reported that Insta360 disclosed more than half of its panoramic camera users now buy these cameras for travel and family records rather than sports, which means the device is moving into ordinary memory work. The user records first and decides less in the moment. The camera keeps surplus context. It reads signals later before turning the day into something shareable. The skill shifts from framing every shot to supervising what the system should keep. The user also has to judge how the system shapes delivery.

Panoramic capture is becoming an automated log, not a format

Panoramic video used to sound like a format choice: record a sphere, then decide later where the viewer should look. Insta360 X6 makes it closer to a daily log. According to ifanr, Insta360 says more than half of its panoramic camera users now buy these cameras for travel and family records rather than sports, which changes the product problem. The camera is no longer only chasing a rider. It also has to serve people far from slopes or dive sites. It has to survive ordinary days when the user may not want to direct every shot.

The hardware still matters. Geekpark says X6 uses a 1/1.1-inch square sensor better suited to panoramic imaging, and ifanr says that sensor was jointly customized with Sony. 36Kr reports support for up to 8K 50 fps panoramic video and 120 million-pixel panoramic photos. Those specs give the software more room to crop after capture. They also let the system pan later while stabilization works without asking the user to frame the scene at capture time.

That is the practical reframing.

AI Director is the clearest sign of the shift. Geekpark says the feature automatically analyzes the previous natural day's footage every night, edits it, then pushes the finished video to the user's phone, without manual upload or manual selection. Ifanr describes a similar charging-time routine: analyze a day's footage, select clips, create camera movements, then edit the video while the device is charging.

This does not remove control; it moves the control point. Insta360 says PanoMind AI can identify highlight moments. It can also design virtual camera movement trajectories while assessing how editable each piece of footage is, according to ifanr. 36Kr adds that gimbal mode 2.0 can directly output 4K flat video while retaining 8K panoramic footage. The useful promise is simple: keep the full scene, receive a flat clip, then review what the machine chose.

Attention becomes an input the camera can read

The operator skill around Insta360 X6 is moving from holding a frame to leaving readable traces for the camera. Geekpark says the X6 can use a head-tracking module first launched with the Luna gimbal camera series, and its basic logic is simple: the camera follows the user's head movement. On X6, geekpark claims head turns and pauses can become attention input for AI Director.

That changes what a "shot" is.

If the camera is worn on the chest, ifanr says the first-person head-tracking module can output flat video from the user's head direction. 36kr describes the same pairing as automatic direction changes based on line of sight. The user is no longer only pointing the camera by hand. A look can become an instruction the system can read. So can a pause before the user continues.

Manual control does not disappear. Ifanr notes that gimbal mode 2.0 can create a virtual gimbal inside a 360° image, with camera movement controlled through a joystick selfie stick. Zoom can be controlled there too. The system can then output 4K flat video. That still gives the operator a way to steer. But the steering sits beside automatic tracking that 36kr says covers people. It also covers animals. Objects are included as well.

The other signal is language. 36kr says X6's Time Pro lets users tell the system what kind of video they want, after which AI selects footage and finishes the video. Ifanr adds that the app includes One-Tap Edit 2.0 and Moments Pro, where users can specify people. They can also specify scenes. Themes are another input, as is video length. The practical craft is therefore becoming supervisory: define who matters, what scene counts, which style is acceptable, and when the camera agent should be trusted. Geekpark reports that Liu Jingkang frames Insta360's "Cameraman" vision as a "Camera Agent."

The edit is split between the edge and the cloud

AI Director makes the X6 feel less like an editing app attached to a camera and more like a scheduled worker inside the camera. According to geekpark, it analyzes the previous natural day's footage every night, edits automatically, and pushes the finished video to the user's phone. It also runs locally and does not require an internet connection. The catch is operational, not conceptual: the panoramic camera needs to be plugged into a power cable for this workflow.

That charging requirement says where the boundary is.

The X6 has enough local compute to do meaningful post-capture work. geekpark says the camera contains 3 chips, including one 4nm SoC in the same class as flagship smartphones and two dedicated imaging chips. ifanr reports its computing power reaches 5 times that of the previous generation, and that battery life reaches 140 minutes at 8K 30 fps, which is 51% higher than X5. Even so, geekpark says the device can produce instant shooting output but cannot edit while shooting when spherical stitching, high-pixel throughput, and on-device inference are happening at the same time.

So the local promise is clear but bounded: while the device is charging, AI Director can analyze a day's footage, select clips, create camera movements, and edit video, according to ifanr. During shooting, the camera still has to prioritize capture and processing. This matters because an outcome-directed camera only works if users understand when the agent is active. Offline automation is useful; hidden waiting states are not.

The cloud path adds another contract. X6 users can still use the mobile app's existing one-tap video creation feature, and they can choose Cloud Moments Pro, which invokes a cloud model. Liu Jingkang told geekpark that Insta360 has tried providing AI editing directly through the cloud. That service is priced at 6 yuan per video, completed hundreds of thousands of edits during the first two months of grayscale testing, and had an export rate above 50%.

The unresolved product question is not whether AI can make a clip. It is what runs offline, what leaves the device, what costs extra, and how repeatably creators can correct the cut when the system guesses wrong. Liu also framed the tension plainly to geekpark: cloud generation is asynchronous, while instant capture and output is a basic need for a camera.

From capture to outcome-directed recording: X6, Xmax X2.0, and Cameraman compared

Real-time video models expose the camera's current limits

Xmax X2.0 shows what a camera agent wants to become when latency stops being a back-office problem. According to woshipm, Xmax AI's general-purpose real-time interactive video model can re-render a stream as it plays. CharX swaps a person. ClothX changes clothing. VibeX changes style while preserving motion and lighting at the frame level. MoX turns a drag on a touch screen into animation for a static subject. DimX places virtual characters through a camera view or anchors them to real-world objects.

That is a different contract from Insta360 X6.

The X6 can produce instant shooting output, according to geekpark, but it cannot edit while shooting when spherical stitching happens alongside high-pixel throughput. On-device inference is another constraint.

The timing gap matters because interaction has thresholds. Woshipm cites Jakob Nielsen's usability markers: within 100 milliseconds feels instantaneous. At 1 second, thought begins to be interrupted. After more than 10 seconds, attention is lost. Xmax X2.0 is built around that psychology. Woshipm says its frame-level autoregressive DiT architecture generates and plays video at the same time, and that Xmax X2.0 reached 960p and 24fps four months after Xmax X1's 480p baseline.

The X6 is closer to a capture agent than a live video model. It records enough surrounding reality to let software decide later; Xmax X2.0 changes what the viewer sees during the act itself. Free mode makes that direction explicit, using text to control effects such as laser eyes or a body catching fire and becoming a skeleton.

Camera hardware still has hard borders. Xmax X2.0 can run local streaming inference at 384@16fps on the latest iPhone, according to woshipm, and its API price is 20 yuan per hour. Insta360 X6 has 140 minutes of battery life at 8K 30 fps, which is 51% higher than X5, according to ifanr. Those figures explain the near future: more immediacy, but still negotiated against stitching and inference. Battery life remains a limit. Heat also shapes what can run live. Liu Jingkang's point, reported by geekpark, is the useful summary: cloud generation is asynchronous, while instant capture and output is a basic need for a camera.

When outcome-directed recording helps-and when to stay manual

• You want a travel or family record without reviewing hours of footage yourself. Use Insta360 X6's AI Director workflow: it analyzes the previous natural day's footage every night, edits it automatically, and pushes the finished video to the user's phone. The device must be plugged into a power cable, and the process runs locally without requiring an internet connection, manual upload, or manual clip selection.
• You know the subject, scene, theme, or desired length and want the system to assemble the video around those constraints. Use the X6 app's Moments Pro, which lets users specify people, scenes, themes, and video length. If you want natural-language direction, the X6's Time Pro feature is described as letting users tell the system what kind of video they want, after which AI selects footage and completes the finished video.
• You need flat video from 360° footage but still want to keep the panoramic master for later reframing. Use Insta360 X6 gimbal mode 2.0: it can directly output 4K flat video while retaining 8K panoramic footage, and it supports camera movement, zoom, and tracking with a joystick selfie stick.
• Your attention or line of sight is the best signal for what matters in the scene. Use the X6 with the head-tracking accessory. When paired with the accessory, the camera can automatically change its shooting direction based on the user's line of sight; the head-tracking module's core logic is that the camera follows where the user's head moves.
• You need real-time character, clothing, or style transformation rather than faithful capture. Use Xmax X2.0 rather than a capture-first camera workflow. It provides real-time re-rendering capabilities including CharX real-time person replacement, ClothX real-time clothing replacement, and VibeX real-time style replacement, plus Free mode for text-controlled effects.
• You are trying to reconstruct a location as a navigable space instead of producing a conventional clip. Try Insta360's 3D Time Capsule, which uses 3D Gaussian splatting to reconstruct a captured scene so users can move freely and revisit the shooting location from different positions. Do not treat it as mature for every environment: the source describes current 3D Gaussian splatting as mainly aimed at static spaces, and 3D Time Capsule as still an early trial-stage feature.

Spatial reconstruction is useful only if it fits daily output

3D Time Capsule pushes the X6 idea past reframing video. According to geekpark, Insta360 released the app feature alongside X6, and it uses 3D Gaussian splatting to reconstruct a captured scene. The result is not a flat clip with a chosen crop. Users can move through the generated scene and revisit the shooting location from different positions.

That matters because it treats recording as capture for later navigation, not only later editing.

The limit is also clear. Geekpark describes current 3D Gaussian splatting as mainly aimed at static spaces, which makes 3D Time Capsule an early trial-stage feature rather than a default format for daily creators. A freely viewable room or street corner is useful when the output needs inspection. It can also serve memory. It is useful when the creator needs a virtual camera move. It is less obviously useful when the target is a short social post watched in a feed.

This is where spatial reconstruction has to meet the same test as Time Pro. According to 36kr, Time Pro lets users describe the kind of video they want in natural language, after which AI selects footage and completes the finished video. 36kr also says X6 adds 3D spatial reconstruction that converts captured footage into freely viewable 3D scenes. The connection is practical: reconstruction becomes valuable when the system can turn navigable space into a finished result, rather than asking the creator to wander through another interface.

Woshipm's Xmax X2.0 examples show the adjacent direction: CharX replaces a person in real time, ClothX changes clothing, VibeX changes style, and DimX can place virtual characters through a camera or anchor them to real-world objects. Free mode goes further with text-driven effects such as laser eyes or a body catching fire and becoming a skeleton.

For everyday output, the question is not whether spatial scenes are impressive. It is whether they shorten the path from captured reality to an acceptable clip. Until spatial displays are common, the useful skill is deciding when to preserve a place as navigable space, and when to let AI collapse it back into video.

Treat the next camera decision as a workflow test, not a spec contest. For a trip, match, family day, or event, run one pass where the device simply records from a mount or wearable position, then judge the first AI cut against what you would have edited by hand. Watch the weak spots Liu Jingkang flagged to geekpark: cloud generation is still asynchronous, while a camera's basic promise is instant capture and output. Ultra-long telephoto detail is also a boundary, not a magic fix.

Keep an eye on form factor. Liu said Insta360's gimbal technology began in conference cameras, its AI noise reduction work started in 2019, and Cameraman already exists in drone form. Phones, glasses, thumb cameras, head-mounted action cameras, wearable devices, and drones will ask for different signals from you.

For readers outside China

• Availability: Insta360 X6 was released in mainland China, where it starts at 3,999 yuan. The source material does not say whether X6, AI Director, 3D Time Capsule, Moments Pro, or the head-tracking accessory are available outside China. Xmax X2.0 was officially announced on X and opened its API at the same time, but the source material does not specify country-by-country availability.
• Pricing: Insta360 X6 starts at 3,999 yuan in mainland China. Insta360's cloud AI editing service is priced at 6 yuan per video. Xmax X2.0's API price is 20 yuan per hour. Other prices, subscription terms, accessory prices, and international pricing are not disclosed in sources.
• Closest Western equivalents: Insta360 X6 is closest to a 360° action camera workflow such as GoPro Max or other panoramic action cameras, but with stronger emphasis on automatic editing, local AI Director, and AI-assisted reframing.; X6 gimbal mode 2.0 resembles software reframing or virtual-gimbal workflows used with 360° footage, where a flat 4K output is chosen from a larger spherical capture.; 3D Time Capsule is closest to consumer-facing Gaussian-splatting or photogrammetry-style spatial reconstruction tools, but the sources describe it specifically as using 3D Gaussian splatting.; Xmax X2.0 is closest to real-time video effects, AR filters, and generative video re-rendering tools, but its source positioning is a real-time interactive video model with API access.
• Data residency: Insta360 X6's AI Director runs locally on the device and does not require an internet connection, manual upload, or manual footage selection. X6 users can also choose Cloud Moments Pro mode, which invokes a cloud model, and Insta360 has tried cloud AI editing services, but the sources do not say where that cloud processing happens or where data is stored. Xmax X2.0 has an API and is described as deployed on mobile devices with local streaming inference at 384@16fps on the latest iPhone, but data residency, retention, and hosting locations are not disclosed in sources.

Sources

• woshipm 为什么说以Xmax X2.0为代表的实时交互视频模型值得重视？ https://woshipm.com/ai/6430819.html
• geekpark 这家靠全景起步的公司，正在逐步让用户「忘掉相机」 https://geekpark.net/news/368847
• ifanr 对话影石创始人刘靖康：全景相机的新范式，是「空间相机」 https://ifanr.com/1675029
• geekpark 对话刘靖康：镜头之外的「智能」和「审美」，才是 Camera Agent 的主战场 https://geekpark.net/news/368503
• 36kr 最前线｜3999元起，影石发布X6：全景相机开始把更多创作交给AI https://36kr.com/p/3936487630650758