Video · image-to-video · video-generation

MAGI-1

sand-ai

Added as a high-priority open video generation candidate with Apache-2.0 HF metadata.

IntegratedDedicated environmentPinned sourcemagi-1

Build your run

Choose a recorded variant, then copy the exact setup or inspection command.

worldfoundry.pipeline

Variant

Taskimage-to-video

Environmentmagi-official-cu124

DeviceCUDA 12.4

worldfoundry-eval evaluate \
  --mode model \
  --model-id magi-1 \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/magi-1 \
  --metric artifact_count \
  --json

Compatibility & versions

Manifest-backed recipe

Integration: integrated
Runner evidence: Not recorded
Environment: magi-official-cu124
Python: 3.10.12
CUDA: CUDA 12.4
PyTorch: —
Source revision: 6ff822e74ded50611e81e1d0e115146b5c4dd2a5
Checkpoint revision: cd0a46589f3279a3c5c579f5f3c693a6ad36d0d4

Runtime profile: magi-1
Pipeline binding: magi-1
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.video_official.pipeline_official_video:MAGI1Pipeline
Backend stage: official_runtime_bridge
Runtime status: integrated_checkpoint_required
Driver status: compatible
Environment kind: Dedicated environment

Install environment

The environment resolver reads the recorded profile and chooses the unified or dedicated environment shown here.

bash scripts/setup/model_env_install.sh --model magi-1

Package constraints 18

accelerate==0.32.1
beautifulsoup4==4.13.4
debugpy==1.8.14
diffusers==0.29.2
einops>=0.6.0
ffmpeg-python
ftfy==6.2.0
gpustat==1.1.1
imageio==2.34.0
imageio[ffmpeg]
matplotlib==3.10.1
numpy==1.26.4
protobuf==5.28.3
rich==14.0.0
sentencepiece==0.2.0
timm==1.0.15
torchdiffeq==0.2.4
transformers==4.42.3

Conda packages 7

python=3.10.12
pip
ffmpeg=4.4
pytorch==2.4.0
torchvision==0.19.0
torchaudio==2.4.0
pytorch-cuda=12.4

Checkpoints & assets

Run the local check before allocating compute. Gated, private, and license fields below come directly from the checkpoint manifest.

worldfoundry-eval zoo model-download --model-id magi-1 --check-local --json

sand-ai/MAGI-1

Revision: cd0a46589f3279a3c5c579f5f3c693a6ad36d0d4
License: apache-2.0
Gated: false
Private: false

sand-ai/MAGI-1

Revision: —
License: —
Gated: —
Private: —

Launch & outputs

The generated command uses the shared evaluation boundary and writes durable result manifests and artifacts.

worldfoundry-eval evaluate \
  --mode model \
  --model-id magi-1 \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/magi-1 \
  --metric artifact_count \
  --json

Input contract

Field	Recorded contract
`prompt`	Required
`image`	Required

Artifact contract

Artifact kind	Filename / path
`generated_video`	`magi-1.mp4`

Evidence & sources

Catalog integration, native-demo parity, and runner parity are independent records.

Integration: integrated
Runner evidence: Not recorded
Native demo evidence: pending
Validation imports: torch, torchvision, flash_attn, flashinfer, diffusers, transformers, sentencepiece, torchdiffeq

Recipe provenance

worldfoundry/data/models/catalog/video/magi-1.yamlGitHub6ff822e74ded50611e81e1d0e115146b5c4dd2a5

Build your run

Input contract

Artifact contract

Related recipes