Xmax X2.1 transforms characters and sets while the camera shoots

Xmax X2.1 announces live video transformation up to 1080p, featuring character, clothing, and background swaps at $0.0016 per second.

Changing a character's face during a live broadcast, instantly replacing their clothes, or completely transforming the background without interrupting the video. This is the type of interaction offered by Xmax X2.1, the new version of the real-time video generation model developed by Xmax AI. The company is announcing an improvement in visual quality up to 1080p, with no increase in the displayed price, set at $0.0016 per second.

Unlike traditional video generators, which produce sequences after several seconds or minutes of processing, Xmax seeks to transform a video stream while it is being broadcast. Incoming frames are processed continuously, with the ability to modify characters, objects, and the appearance of the environment using prompts or visual references.

This approach is particularly aimed at streaming platforms, social experiences, e-commerce, and interactive applications. Live video generation that follows movement

The main difference with Xmax lies in its continuous operation. The model does not simply create a clip from a prompt: it receives a stream from a camera or a video and generates a new frame as the scene unfolds.

The goal is to maintain the continuity of gestures, expressions, and movements, even when the subject's appearance is profoundly modified.

In the company's demonstrations, a person can be replaced by another character while retaining their body movements. Their expressions, gaze, and gestures serve as the foundation for the new representation.

The system can also modify the environment around the subject, preserving the relationship between camera movements and elements present in the scene.

Xmax refers to this as re-rendering: the video stream becomes a source of movement and visual information that the model continuously reinterprets.

This distinction brings the technology closer to video filters and augmented reality systems, but with much more significant generative transformations than simple overlaid effects. Separately modifiable faces, clothing, and hairstyles

With X2.1, Xmax claims greater precision in selecting elements to transform.

It becomes possible to replace an entire character, but also to adjust more specific features: face, head, hairstyle, clothing, colors, or background.

Changing outfits is one of the highlighted use cases. Using a visual reference, the system seeks to apply a garment to a filmed person, while maintaining their posture, movements, and the scene's lighting conditions.

This feature is of particular interest to e-commerce, where a model could present multiple outfits without multiplying shoots.

It can also be used for virtual try-on, allowing a user to visualize a garment on their own body.

However, the demonstrations do not allow for establishing the exact fidelity of how fabrics drape, or the accuracy of sizes and cuts. A convincing video transformation does not necessarily equate to a textile simulation precise enough to guarantee the actual real-life rendering of a product. Visual effects modified during the broadcast

Xmax X2.1 also extends its capabilities to generative visual effects.

A user can add, remove, or replace certain elements of a scene, modify its atmosphere, or completely transform its aesthetic.

The company illustrates several categories of interactions within its software environment.

CharX involves character replacement, ClothX handles clothing changes, and VibeX transforms the overall visual style.

Other features, such as DimX, allow for the introduction of virtual characters into a real environment. MoX relies on touch interactions: the user can move or animate certain elements by drawing a path directly on the screen.

These features build upon the capabilities previously introduced with Xmax X2.0. The new version is primarily advertised as an improvement in quality, realism, and prompt adherence.

In its launch communication, the company has not published an independent benchmark quantifying X2.1's progress across these various criteria. Announced 1080p with no price increase

Xmax highlights a 1080p output resolution, at a time when interactive video generation experiences typically have to deal with significant compute and transmission constraints.

The claimed price remains fixed at $0.0016 per second, which equals roughly $0.096 per minute and $5.76 per hour of continuous generation, excluding potential additional fees.

Per second · Per minute · Per hour $0.0016 · $0.096 · $5.76

Amounts calculated from the price announced by Xmax, for a continuously billed session.

One point is worth distinguishing, however: the publicly available documentation still presents settings and limits corresponding to X2.0, notably a target resolution of approximately 1.28 million pixels per frame. Actual 1080p support by X2.1, depending on the format and streaming mode, therefore remains to be confirmed in updated technical specifications.

The same applies to end-to-end latency. The documentation explains that latency depends not only on the model's processing, but also on the uplink and downlink connections used to transmit the video. No independent results yet guarantee identical latency across all networks and devices. A proprietary model rather than an assembly of video services

Xmax positions itself as a developer of specialized models for interactive video transformation. Its previous versions rely on a generative infrastructure capable of processing frames from a stream continuously.

The company has its own demo platform, APIs, and several SDKs allowing the technology to be integrated into existing applications.

The JavaScript SDK published on GitHub documents camera stream acquisition, transformation prompts, reference images, and the generated video output. Android integrations are also available.

The SDK documentation explicitly refers to model identifiers from the X2.1 branch, confirming its technical integration beyond just social media announcements.

However, the model weights and a full scientific description of its architecture are