# PKU-YuanGroup/Helios

Helios: Real Real-Time Long Video Generation Model

Repository: https://github.com/PKU-YuanGroup/Helios
Canonical: https://ross.abutalabs.com/products/pku-yuangroup-helios
Homepage: https://pku-yuangroup.github.io/Helios-Page
Language: Python
License: Apache-2.0
License Family: permissive
Topics: acceleration, diffusion, diffusion-models, image-to-video, long-video-generation, real-time, text-to-video, video-generation, video-generator, video-to-video, world-model, world-models, diffusion-model, efficient-tuning, interactive, long-context, image2video, text2video, video2video, high-quality
Last push: 2026-08-24T07:26:04+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 35, longevity 13
- inputs: {"age_days": 184, "days_push": 9, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2076, forks 166 (observed 2026-08-28T04:06:11.226623+00:00)

## What it is
Helios is a 14B autoregressive diffusion model for real-time, minute-scale video generation supporting text-to-video, image-to-video, and video-to-video tasks. It achieves 19.5 FPS on a single H100 GPU without conventional anti-drifting heuristics or standard acceleration techniques.

## Use cases
- generate long videos in real time from a text prompt
- turn a single image into a minute-long video
- real-time interactive video generation on a single GPU
- video-to-video restyling with a diffusion model
- build an interactive world model that streams frames
- research on long-video coherence without anti-drifting tricks

## When to choose
- you need high-quality minute-scale video generation at real-time frame rates
- you want a single 14B model handling T2V, I2V, and V2V
- you have an H100 GPU or Ascend NPU and want efficient large-model inference
- you are researching autoregressive diffusion video models

## When to avoid
- you only need lightweight or CPU-based video generation
- you lack a high-end GPU (H100-class) for real-time performance
- you need a turnkey end-user video app rather than a model and codebase

## Facets
- artifact type: library
- maturity: active
- function: video-processing, machine-learning, deep-learning, llm-inference
- domain: deep-learning, artificial-intelligence, gpu-computing
- platform: python
- tags: diffusion-models, text-to-video, image-to-video, video-to-video, real-time-generation, world-model, autoregressive, video-generation, video, gpu, linux

## Member repositories
- PKU-YuanGroup/Helios (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:11.226623+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:56:10.697480+00:00, confidence not recorded.
  - readme: https://github.com/PKU-YuanGroup/Helios (fetched 2026-08-28T04:06:11.226623+00:00, sha 89f206cf8961)
  - homepage: https://pku-yuangroup.github.io/Helios-Page (fetched 2026-08-29T10:36:25.088252+00:00, sha d12bdc902d16)
- Data as of 2026-08-30T08:39:29.467469+00:00.