HyperFlow for MiniMax-H3
HyperFlow for MiniMax-H3 from Video Rebirth: text-to-video and image-to-video; Runner parity pending; dedicated `hyperflow`.
Eight-step MiniMax-H3 adapter with joint video/audio, first/last-frame conditioning and ordered multimodal references.
No official teaser or method figure is checked into this docs tree.
Code: Video-Rebirth/hyperflow.
Build your run
HyperFlowPipelineworldfoundry.pipelines.hyperflow.pipeline_hyperflowworldfoundry.pipelineworldfoundry-eval run \
hyperflow \
--pipeline.task-profile t2va \
--jsonWhat it is for
Use hyperflow for the family's joint T2V/I2V inference path. The catalog records both task profiles on this page.
It is published by Video Rebirth.
Catalog tasks include text-to-video, image-to-video, reference-to-video, and audio-video-generation. Catalog aliases include hyperflow-h3 and minimax-h3-hyperflow. Documented modalities: text and image in, video out.
Runtime contract
Optional prompt, num_frames, height, width, and seed; artifacts generated_video and hyperflow.mp4.
CLI task profiles: t2va, fl2va, and ref2va. WorldFoundry binds HyperFlowPipeline.
Family status: Runner parity pending.
Recorded t2va defaults: seed=42 Recorded fl2va defaults: seed=42.
Supply a complete local diffusers MiniMax-H3 snapshot, weights_path and the dedicated Python executable. The upstream loader supplies the fixed grid; num_inference_steps must be 8 and output fps must be 24.
Running in WorldFoundry
Launch worldfoundry-eval run hyperflow --pipeline.task-profile t2va to sample the clip. The command builder above switches --pipeline.task-profile among recorded tasks and still emits worldfoundry-eval run.
Recorded weight repositories include videorebirth/hyperflow and MiniMaxAI/MiniMax-H3. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh hyperflow. Add --download to fetch missing public checkpoints.
Status: Runner parity pending.
