WorldFoundry
English中文

SAMA

Method overview: Overall pipeline. SAMA first performs factorized pre-training (stage 0) on additional perturbed videos by completing a pretext task conditioned on the given captions. It then performs normal supervised fine-tuning.

SAMA: video-editing; Runner parity pending.

Edit this pageGitHubMarkdown

SAMA 14B is an open instruction-guided video editing model: it takes a source video and a text instruction. The released inference route does not accept a spatial mask or separate reference image. WorldFoundry reports missing checkpoints at runtime; weights are apache-2.0 at syxbb/SAMA-14B.

Method overview: Overall pipeline. SAMA first performs factorized pre-training (stage 0) on additional perturbed videos by completing a pretext task conditioned on the given captions. It then performs normal supervised fine-tuning.
Method overview: Overall pipeline. SAMA first performs factorized pre-training (stage 0) on additional perturbed videos by completing a pretext task conditioned on the given captions. It then performs normal supervised fine-tuning.

The recorded paper is arXiv 2603.19228.

Project page: cynthiazxy123.github.io/SAMA. Code: Cynthiazxy123/SAMA.

Build your run

Pipeline
SAMA14BPipelineworldfoundry.pipelines.video_official.pipeline_official_video
TaskInteractive Video interactive-video
Environmentworldfoundry-unified-cu128
DeviceCUDA 12.8
Runnerworldfoundry.pipeline
Loadingfrom_pretrained
Invocationnative_or_standard_invocation
worldfoundry-eval run \
  sama-14b \
  --pipeline.task-profile interactive-video \
  --json

What it is for

Use sama-14b for catalog video-to-video inference. Start here when the input is already a video.

The catalog records a 14B-parameter editing model with a prompt/image/video input contract; the publishing institution and architecture internals are not recorded in this repository — consult the upstream GitHub (Cynthiazxy123/SAMA) for details.

Catalog tasks include video-editing.

Runtime contract

Optional prompt, negative_prompt, input_path, frames, and fps; artifacts sama-14b.mp4.

CLI task profiles: interactive-video. WorldFoundry binds SAMA14BPipeline.

Family status: Runner parity pending.

Recorded interactive-video defaults: fps=16, height=480, width=832, seed=1.

Supply a source video and edit instruction. A checkpoint-backed 17-frame run was compared with the official implementation; broader editing quality remains unverified.

Running in WorldFoundry

Launch worldfoundry-eval run sama-14b --pipeline.task-profile interactive-video to edit the clip. The command builder above switches --pipeline.task-profile among recorded tasks and still emits worldfoundry-eval run.

It runs in the unified environment worldfoundry-unified-cu128 (Python 3.11, CUDA 12.8).

Recorded weight repositories include syxbb/SAMA-14B. Prepare the environment and check local assets with bash scripts/inference/prepare_model_infer.sh sama-14b. Add --download to fetch missing public checkpoints.

Recorded licenses include apache-2.0; status: Runner parity pending.