Twelve references to guide video generation with BACH 1.0 Pro

Video Rebirth launches BACH 1.0 Pro, an AI video model capable of combining 12 visual and audio references for productions up to 4K.

Generating a beautiful shot is not enough to make a commercial or a film. You need to maintain the same characters, preserve product details, respect brand identity, and keep consistency across sequences. BACH 1.0 Pro, the new video model from Video Rebirth, tackles these production constraints with a multimodal reference system. Images, videos, and audio files can serve as the starting point for a generation, while the model already holds second place on an image-to-video leaderboard by Artificial Analysis. Twelve references to build a scene

The main innovation of BACH 1.0 Pro lies in its announced ability to leverage up to 12 references in a single generation.

Creators can combine character images, product photographs, video clips, or audio excerpts to guide the output. These elements serve to define appearances, camera movements, pacing, or the sound atmosphere.

The goal is notably to prevent a character from changing faces, a product from losing certain details, or brand-specific graphic elements from being distorted from one frame to the next.

This workflow is of particular interest to commercial productions and films using recurring characters. It allows users to provide concrete references to the model rather than relying solely on a text description.

Video Rebirth claims improved retention of visual identity, though it has not published specific success rates for generations using all twelve references. Three generation modes, a fourth in development

BACH 1.0 Pro offers several ways to build a sequence.

The text-to-video mode generates a video from a description. The image-to-video mode allows users to input an initial image, a final image, or both, in order to generate a transition between defined compositions.

The third mode, reference-to-video, simultaneously leverages different media to guide generation. It stands as one of the model's main selling points for professional projects, particularly when character identity, movements, and visual continuity must be controlled.

A fourth mode, video-to-video, has been announced but is not yet available. It is intended to allow editing an existing video based on a text prompt while preserving its movement structure and camera pans.

The distinction is important: BACH currently offers guided generation tools, but its direct video modification system is yet to come. Second place at Artificial Analysis

On Artificial Analysis's AA-Video-I2V-Silent v1.0 leaderboard, dedicated to image-to-video generation without audio evaluation, BACH 1.0 Pro ranks second with 1,362 Elo points.

It follows Gemini Omni Flash, credited with 1,368 points, and is tied in score with Wan 3.0.

This ranking is based on blind side-by-side video comparisons. The Elo score reflects observed preferences between different generations, rather than an absolute measure of visual fidelity or cinematic quality.

Nor does it allow for the conclusion that BACH systematically outperforms its competitors on character continuity, reference accuracy, or multi-shot editing. Generation in 1080p and 4K

BACH 1.0 Pro is available via a developer API, offering two resolution tiers.

Generation costs $0.12 per second in 1080p and $0.24 per second in 4K.

A ten-second video therefore costs $1.20 in 1080p or $2.40 in 4K, without additional video references.

Images and audio files used as references incur no extra charge. However, when a video sequence is used as a reference, its duration is billed additionally based on the output resolution.

For example, a ten-second video generated in 1080p using five seconds of video reference costs $1.80.

References must be accessible via public HTTP or HTTPS URLs. The API does not directly support Base64-encoded files, a constraint to anticipate for companies handling confidential content. Video generation closes in on production constraints

Video Rebirth presents BACH 1.0 Pro as an engine designed for film, advertising, e-commerce, and animation.

Its positioning reflects changing expectations among creators. The quality of a single generation still matters, but the ability to repeat a character, preserve a product, and produce multiple shots belonging to the same universe is becoming just as crucial.

BACH's promised control does not eliminate the need for art direction, take selection, and post-production. Results still depend on the provided references and the model's ability to interpret them correctly.

The question is no longer just whether an AI can create a convincing video, but whether it can reproduce a visual intent faithfully enough to fit into an actual production pipeline. This is the difference on which Video Rebirth intends to position BACH 1.0 Pro.