Train your own LoRA
A Civitai alternative for training your own LoRA
On Root you train a LoRA on your own pictures, then use it in video and image models in the same project. Three of the LoRAs you can train work in Root's models: MiniMax H3 LoRA, Krea 2 LoRA and Wan 2.2 video LoRA. Training needs the Indie plan ($49/mo) or above; it isn't part of the free month. With the default settings, a training run costs about $1.85 to $18.80.
Updated
First month free on the Beginner plan, with 650 credits to try any model.
Made in Root
LoRAs you can use in Root's models
Each one trains on a set of pictures of a character, a product or a style. Prices are for one training run with the default settings, in dollars. 100 credits is $1.
| What you train | Use it in | Training |
|---|---|---|
| MiniMax H3 LoRA | MiniMax H3 text or image to video, MiniMax H3 reference to video and the MiniMax H3 swap | about $1.85 |
| Krea 2 LoRA | Krea 2 (custom LoRA) images | about $4.61 |
| Wan 2.2 video LoRA | Wan 2.2 text to video, Wan 2.2 VACE restyle, Wan Animate and Wan Animate 2 | about $18.80 |
Prices marked “about” are for work that runs on Root's own machines: the price shown before the run is an estimate, and the bill is set by how long the run really takes.
How a trained LoRA works in Root
- It is saved to your workspace's models and can be picked in every project there.
- It shows up only on the models it fits, so a video LoRA is never offered on an image model.
- You can use up to 2 LoRAs in one run, each with its own strength.
- Every clip and image you make with it is kept in your project, with the rest of your takes.
- The Indie plan keeps up to 2 trained models; bigger plans keep more.
How to train and use a LoRA in Root
- Build a dataset. In Models, make a dataset: upload pictures of the character, product or style, or add ones from your projects, and caption them.
- Start a training run. In Train a new model, pick the dataset and what to train, and name it. The estimate shows before you start.
- Wait for it to finish. Training runs on Root's own machines. The finished model can be picked in every project.
- Use it in a shot. Add a node with a model that takes your LoRA, pick the LoRA, and write the shot as usual.
Related guides
LoRA training questions
What is a good Civitai alternative for training a LoRA?
Root. You train a LoRA on your own pictures and use it in MiniMax H3 and Wan 2.2 video and Krea 2 images, in the same project. Training needs the Indie plan ($49/mo) or above; it isn't part of the free month. With the default settings, a run is about $1.85 to $18.80.
How much does it cost to train a LoRA on Root?
With the default settings, MiniMax H3 LoRA is about $1.85, Krea 2 LoRA is about $4.61 and Wan 2.2 video LoRA is about $18.80. Training is billed on how long it really takes, so the figure shown before you start is an estimate.
Can I train a LoRA for video?
Yes. Two of the three in the table above are for video: MiniMax H3 LoRA and Wan 2.2 video LoRA. The table shows the models each one works in.
Do I need a paid plan to train a LoRA?
Yes. Training needs the Indie plan ($49/mo) or above; it isn't part of the free month.
Can I use more than one LoRA at once?
Yes, up to 2 in one run, each with its own strength.
Who can use the LoRAs I train?
Your workspace. A model you train is saved to your workspace's models, and it can be picked in every project there.
Can I train or use a LoRA by API?
No. Training, and making clips and images with your LoRA, happen in the Root app.
Try Root free for a month
The first month of the Beginner plan is free and comes with 650 credits to spend on any model. After that it's $9 a month, and you can cancel whenever you like.