World models · camera-control-video · text-to-video · video-diffusion-plugin

AC3D

Camera-Control Video

WorldFoundry vendors the official AC3D inference runtime under worldfoundry/synthesis/visual_generation/ac3d/ac3d_runtime; WorldFoundry owns path resolution, command construction, output loading, and pipeline/catalog exposure.

Runner verifiedDedicated environmentPinned sourceac3d

Build your run

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

worldfoundry.pipeline

Variant

Taskcamera-control-video

Environmentac3d

DeviceCUDA 12.4

worldfoundry-eval evaluate \
  --mode model \
  --model-id ac3d \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/ac3d \
  --metric artifact_count \
  --json

Compatibility & versions

Manifest-backed recipe

Integration: integrated
Runner evidence: verified
Environment: ac3d
Python: 3.10
CUDA: CUDA 12.4
PyTorch: torch==2.4.0+cu124
Source revision: fe40a8d2dee163b951e6405d8cf1a2c14ecc68eb
Checkpoint revision: —

Runtime profile: ac3d
Pipeline binding: ac3d
Runner: worldfoundry.pipeline
Pipeline target: worldfoundry.pipelines.ac3d.pipeline_ac3d:AC3DPipeline
Backend stage: official_in_tree_runtime
Runtime status: official_in_tree_runtime_ready
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 ac3d

Package constraints 16

torch==2.4.0+cu124
torchvision==0.19.0+cu124
diffusers==0.32.2
transformers>=4.44.0,<5
accelerate>=0.34.2
safetensors>=0.4.5
sentencepiece>=0.2.0
numpy==1.26.0
opencv-python>=4.10.0.84
imageio==2.34.2
imageio-ffmpeg>=0.5.1
moviepy>=1.0.3
pillow==9.5.0
decord==0.6.0
einops
packaging

Conda packages 3

python
pip
ffmpeg

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 ac3d --check-local --json

No checkpoint repository is recorded in the catalog manifest.

Launch & outputs

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

worldfoundry-eval evaluate \
  --mode model \
  --model-id ac3d \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/ac3d \
  --metric artifact_count \
  --json

Input contract

Field	Recorded contract
`prompt`	Required
`image`	Optional
`video`	Required
`actions`	camera_index, camera_index_range, camera_pose_rows

Artifact contract

Artifact kind	Filename / path
`generated_video`	`ac3d.mp4`

Evidence & sources

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

Integration: integrated
Runner evidence: verified
Native demo evidence: Not recorded
Validation imports: torch, torchvision, diffusers, transformers, decord, einops

Recipe provenance

worldfoundry/data/models/catalog/world_models/ac3d.yamlProjecthttps://snap-research.github.io/ac3d/Paperhttps://arxiv.org/abs/2411.18673GitHubfe40a8d2dee163b951e6405d8cf1a2c14ecc68eb

Build your run

Input contract

Artifact contract

Related recipes