Video · text-to-video · image-to-video · video-to-video · long-video-generation

Helios

BestWishYsh

The inference implementation is vendored in-tree at the pinned source revision.

IntegratedDedicated environmentPinned sourcehelios

Build your run

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

worldfoundry.pipeline
Tasktext-to-video
Environmentworldfoundry-helios-cu128
DeviceCUDA 12.8
worldfoundry-eval evaluate \
  --mode model \
  --model-id helios-distilled \
  --model-runner worldfoundry:pipeline \
  --model-manifest-dir worldfoundry/data/models/catalog \
  --requests-path tmp/requests.jsonl \
  --output-dir tmp/model_eval/helios-distilled \
  --metric artifact_count \
  --json
01

Compatibility & versions

Manifest-backed recipe

Integration
integrated
Runner evidence
pending visual qa
Environment
worldfoundry-helios-cu128
Python
3.11.2
CUDA
CUDA 12.8
PyTorch
torch==2.10.0
Source revision
8f2a2faab3298c8a7630a2c73aea37c01b5bab01
Checkpoint revision
1999182614cb08d3bdcc46b9827504af2914b87b
Runtime profile
helios
Pipeline binding
helios
Runner
worldfoundry.pipeline
Pipeline target
worldfoundry.pipelines.helios.pipeline_helios:HeliosPipeline
Backend stage
official_in_tree_runtime
Runtime status
all_three_checkpoint_variants_workspace_validated_a100
Driver status
compatible
Environment kind
Dedicated environment
02

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 helios-distilled
Package constraints 20
  • pyyaml
  • torch==2.10.0
  • torchvision==0.25.0
  • triton==3.6.0
  • kernels==0.12.1
  • git+https://github.com/huggingface/diffusers.git@b16635e90efd1ac32360efad099d793c10413b35
  • transformers==5.3.0
  • peft==0.18.1
  • accelerate==1.12.0
  • huggingface-hub==1.4.1
  • numpy<2.0.0
  • opencv-python
  • moviepy
  • imageio-ffmpeg
  • video-reader-rs==0.4.1
  • aiohttp>=3.9,<4
  • aiortc>=1.14,<2
  • av>=14,<17
  • ftfy
  • packaging
Conda packages 4
  • python=3.11.2
  • pip
  • ffmpeg
  • cuda-nvcc
03

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 helios --check-local --json
BestWishYsh/Helios-Distilled
Revision
1999182614cb08d3bdcc46b9827504af2914b87b
License
apache-2.0
Gated
Private
BestWishYsh/Helios-Base
Revision
21cf91da9cae0270a9405000c2e798f83770b1dd
License
apache-2.0
Gated
Private
BestWishYsh/Helios-Mid
Revision
23d110fa638959ac2a9aaa024d372519d54d1d2b
License
apache-2.0
Gated
Private
BestWishYsh/Helios-Distilled
Revision
License
Gated
Private
BestWishYsh/Helios-Base
Revision
License
Gated
Private
BestWishYsh/Helios-Mid
Revision
License
Gated
Private
04

Launch & outputs

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

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

Input contract

FieldRecorded contract
promptRequired
imageoptional
videooptional
actionsseed, sample_type, variant, num_frames, enable_low_vram_mode, nproc_per_node, prompt_update

Artifact contract

Artifact kindFilename / path
generated_video
generated_videohelios.mp4
05

Evidence & sources

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

Integration
integrated
Runner evidence
pending visual qa
Native demo evidence
Not recorded
Validation imports
yaml, torch, torchvision, triton, diffusers, transformers, accelerate, cv2