lucataco / ai-toolkit

Ostris AI-Toolkit for Flux LoRA Training (Proof of Concept). Please use the official trainer at: ostris/flux-dev-lora-trainer

  • Public
  • 46.2K runs
  • GitHub
  • Weights
  • License
Run with an API
Playground API Examples Train README Versions

Run time and cost

This model costs approximately $3.08 to run on Replicate, or 0 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia A100 (80GB) GPU hardware. Predictions typically complete within 37 minutes.

Readme

------------------------------------------------

Use the OFFICIAL Flux LoRA Trainer here

It supports more features, runs faster, and is cheaper to use!

------------------------------------------------

About

A Cog implementation of ostris/ai-toolkit

(Training only supports FLUX.1-dev)

Follow me on Twitter/X @lucataco93

How to use

In the TRAIN tab (between README and VERSIONS) you’ll see the parameters that you can select to train a LoRA

For destination select/create an empty Replicate model location to store your LoRAs. (Ex: lucataco/flux-loras)

For images upload your zip/tar file of images for training. Images must be jpg, jpeg, or png. File names must be their captions, ex: a_bird_in_the_style_of_TOK.png, etc. If you include txt files, the text files should be named the same as the images and should contain only the caption.

For model_name use “black-forest-labs/FLUX.1-dev”

For hf_token use your Huggingface token to access the Flux-Dev weights for training. Make sure the Access Token has the right permissions: “Read access to public gated repos you can access”

For steps select a value from 500-4000

By default the trigger word is: ‘TOK’. The other steps are optional

Example Training runs:

Below is an example training run to create a watercolor style LoRA, trained on 16 watercolor images for 1000 steps in 41min: watercolor training

Below is an example training run to create a LoRA of a dog, trained on 26 images of Queso for 500 steps in 21min: queso training

How to test your LoRA

Once you have an Output.zip file you can download and extract the safetensors file, and upload it to a huggingface space (ex: lucataco/flux-dev-lora). If you added a model name for the Train parameter repo_id at the bottom, then this should be done for you automatically

With your LoRA in a huggingface model under your repo_id (lucataco/flux-dev-lora) go to the LoRA Explorer model and try it out. In this example, I trained a watercolor style LoRA, so to activate the LoRA I would use the prompt: “a boat in the style of TOK”

Licensing and commercial use

If you generate images on Replicate with FLUX.1 models and their fine-tunes, then you can use the images commercially.

If you download the weights off Replicate and generate images on your own computer, you can’t use the images commercially.

All Flux-Dev LoRAs have the same license as the original base mode for FLUX.1-dev

If you choose the option to auto-upload your trained LoRA to Huggingface, this License will be added for you