Skip to content
repository radar
Voice, Vision & Multimodal

🎥Goku

FAINT SIGNAL COOLING

Saiyan-World/goku · homepage ↗

[CVPR2025 Highlight] Video Generation Foundation Models: https://saiyan-world.github.io/goku/

⭐ Popular: 3k stars

View on GitHub ↗

repo profile

vintage 2025 1 year old
language Python
license n/a

momentum

total stars 3k
stars added last week +0
commits / week 0 steady
contributors 1
release cadence n/a
last activity 1y 5mo ago

durability

backing foundation-backed academic research (Goku)

since we covered it

monthly average + 44/mo (+2%/mo) · + 805 total since PR#2

why it's a big deal

  • Goku is a joint image-and-video generation foundation model covering text-to-video, image-to-video, and text-to-image in one family, so teams can target several generation tasks from a single approach.
  • The authors report competitive public benchmark scores, including GenEval 0.76 and DPG-Bench 83 for text-to-image and VBench 84 for text-to-video, giving practitioners a documented reference point rather than only claims.
  • The repository accompanies a CVPR 2025 paper from HKU and ByteDance, so it mainly serves researchers tracking rectified flow methods rather than users wanting a ready-to-run tool, since checkpoints and download links are not documented in the README.

under the hood

  • The model is built on rectified flow Transformers, applying a rectified flow formulation to the interaction between video and image tokens.
  • The codebase is entirely Python and is organized into configs, a goku module, and tools, with a requirements.txt for dependencies.
  • The authors attribute results partly to fine-grained image and video data curation alongside the flow-based architecture, though specific model parameter sizes are not stated in the README.

Radar summary, generated from the project's public sources

star history

PR#2 · 2k3k now Feb 2025Aug 2026
  1. PR#2 2k 2025-02-19
  2. now 3k + 805 since first covered

curve is sampled from GitHub's star history, plus our own daily readings since we covered it; the dashed stretch is before we first covered it, the solid line since. figures at coverage are the numbers we printed then (approx.), current count is live.

understory

Better known than its recent output, coasting a little on attention.

-13 understory score output 0 · clout 13
Aug 2025 Jul 2026
  • output, commits & releases
  • clout, star velocity

output = commits & releases; clout = star velocity, both 0 to 100 monthly indices; the gap where output runs above clout is the understory. The understory →

covered in

  • PR#2 2025-02-19 below the radar

    Flow Based Video Generative Foundation Models

similar projects

compare these →
  • 💡 PaddleOCR

    30× the stars

    Turn any PDF or image document into structured data for your AI. A powerful, lightweight OCR toolkit that bridges the gap between images/PDFs and LLMs. Supports 100+ languages.

    88k ACTIVE
  • Unsloth

    25× the stars

    Local UI to run and train LLMs and diffusion models, including Qwen3.8, Kimi K3, MiniMax-H3, Gemma 4, DeepSeek-V4, FLUX and more.

    72k ACTIVE
  • 🎙️ Real-Time-Voice-Cloning

    21× the stars

    Clone a voice in 5 seconds to generate arbitrary speech in real-time

    60k ACTIVE

comments

Sign in with GitHub to add your blip on Goku.

loading comments…