VIVIBIT today announced ongoing testing of MiniMax H3 on the VIVIBIT E1001 Desktop AI Cluster. The setup combines an E1001 shared data platform with four GIGABYTE AI TOP ATOM (DGX Spark) systems powered by GB10 chips. Using ComfyUI, teams can run local AI video workflows while keeping source assets, prompts, workflows, and generated files within their own environment.
MiniMax H3 is a multimodal video-generation model that accepts text, images, video, and audio as inputs. According to MiniMax's published materials, it can generate video with synchronized stereo audio, with support for output up to 2K resolution and 15 seconds. Its multimodal input model makes it relevant to teams creating product videos, branded-character content, concept films, and social assets.
VIVIBIT's test focuses on a practical production question: whether a creative team can repeatedly test ideas, retain the assets behind each version, and keep sensitive material under its own control.
Moving from one-off clips to a repeatable workflow
Cloud video services remain useful for occasional work and elastic capacity, but teams working with unreleased products, talent assets, or brand material often need more control over where files live and how tasks are scheduled. With a local deployment, assets can remain on the team's internal network instead of being uploaded for every generation request. That does not make a workflow automatically secure: teams still need to manage access controls, storage, backups, logs, and network isolation.
MiniMax H3 is publicly available as an open-weight model. It is not an unrestricted open-source release. Organizations should review the , including its commercial-use, territorial, and acceptable-use provisions, before deploying it.
The E1001 serves as the shared asset and workflow layer in the test environment. Teams store source media, model files, and ComfyUI workflows on E1001; GB10 nodes read those inputs and write completed results back to the same location. The result is a single place to review prompt versions, reference assets, intermediate files, and final renders.
The workflow is simple:
Store assets on E1001 → select a ComfyUI workflow → submit the job to an available GB10 node → write the result back to E1001 → review and publish.
Current internal test results
Under VIVIBIT's current test configuration and parameters:
-
A 480P video typically takes about 4 minutes to generate.
-
A 5-second 720P video typically takes about 9 minutes to generate.
These figures are internal test results, not official MiniMax or NVIDIA performance specifications. Resolution, duration, inference steps, model and software versions, reference assets, and workflow configuration can all affect generation time.
The aim is not to claim instant rendering. It is to make the wait productive and predictable. Once a team has refined a prompt and its references, the workflow can be saved in ComfyUI and used again without rebuilding it from scratch. With four independent GB10 nodes, four distinct jobs can run at the same time.
For example, a team can assign one node to a product unboxing video, another to a vertical social-media cut, a third to a branded-character animation, and a fourth to a different prompt or camera treatment. The nodes do not combine to accelerate one individual render; they enable parallel creative work.
Workflows supported by the test environment
The MiniMax H3 workflows being evaluated on E1001 include:
-
Text to video: Create a video from a description of the scene, subjects, action, camera movement, and sound.
-
Image to video: Use a product image, character image, or poster as a visual starting point, then define movement, camera behavior, and audio.
-
First- and last-frame control: Provide the opening and ending frame and generate the transition between them for product transformations, scene changes, logo motion, and shot transitions.
For animation and film development, these workflows can support early motion, shot, and concept testing. For e-commerce and brand teams, they can be used to develop product presentations and multiple delivery formats from the same creative direction. The model does not replace cinematography, editing, or brand review; it gives teams a faster way to test those choices before committing to a final production path.
Cost, privacy, and continuity
The choice between a cloud API and local infrastructure is not simply a price comparison. Cloud services typically charge by duration, resolution, or usage pack. Local deployment eliminates per-clip cloud API charges but introduces hardware, electricity, storage, operations, and update costs. Low-volume or occasional projects may remain a better fit for the cloud. Teams that run frequent iterations, batch jobs, and sensitive-asset workflows may find more value in a local cluster.
The same distinction applies to continuity. Cloud services manage the infrastructure but can impose queues, rate limits, quotas, or service changes. A local environment allows a team to schedule work on its available nodes and retain its workflow templates, while taking responsibility for the equipment and its maintenance.
For current information about MiniMax H3, refer to the . MiniMax's cloud video pricing and terms can change and should be verified directly through the .
Learn more about the .
About VIVIBIT
VIVIBIT develops local AI infrastructure that connects shared all-flash data, high-speed networking, AI nodes, and private AI workflows for small teams.
Media contact
VIVIBIT Communications [pr@vivibit.com]
