Skip to content
repository radar
Models & Inference

🖥️AI-Toolkit

ACTIVE STEADY

ostris/ai-toolkit

The ultimate training toolkit for finetuning diffusion models

⭐ Very popular: 12k stars, gaining about 78 a week

View on GitHub ↗

repo profile

vintage 2023 3 years old
language Python
license MIT

momentum

total stars 12k
stars added last week +78
commits / week 27 steady · +8% vs prior mo
issues closed 88% ~44d to close, median
contributors 29 (+1 in 15d)
release cadence n/a
last activity today

durability

backing community / independent solo maintainer (ostris)
openness permissive

since we covered it

monthly average + 475/mo (+8%/mo) · + 6k total since PR#15

why it's a big deal

  • Consolidates finetuning for a wide range of diffusion models, covering image models like FLUX, SDXL, and Stable Diffusion 1.5 plus video models such as Wan and LTX-2, into one training suite.
  • Targets consumer-grade hardware, so people without datacenter GPUs can train LoRA and LoKr adapters on their own machines.
  • Offers both a CLI and a web UI for job management, which suits both scripted workflows and users who prefer not to run everything from the terminal.

under the hood

  • Written mostly in Python with a TypeScript web interface that runs on Node.js at port 8675, with optional auth via the AI_TOOLKIT_AUTH environment variable.
  • Jobs are defined in YAML configs copied from config/examples and run with python run.py, producing timestamped output folders that store checkpoints and resume training after interruption.
  • Supports LoRA and LoKr training with configurable rank, alpha, and factor, plus layer-specific targeting through only_if_contains and ignore_if_contains network parameters.

Radar summary, generated from the project's public sources

star history

PR#15 · 6k12k now Jul 2023Aug 2026
  1. PR#15 6k 2025-08-20
  2. now 12k + 6k 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

Quietly building: more output than attention, for now.

+14 understory score output 74 · clout 60
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#15 2025-08-20 below the radar

    Diffusion Model Finetuning Suite

similar projects

compare these →
  • 🖥️ Open WebUI

    13× the stars

    User-friendly AI Interface (Supports Ollama, OpenAI API,...)

    149k ACTIVE
  • 🧪 autoresearch

    8× the stars

    AI agents running research on single-GPU nanochat training automatically

    94k ACTIVE
  • 🦙 LLaMA-Factory

    6.3× the stars

    Unified Efficient Fine-Tuning of 100+ LLMs & VLMs (ACL 2024)

    74k ACTIVE

comments

Sign in with GitHub to add your blip on AI-Toolkit.

loading comments…