Skip to content
Skip to workspace
Opening your workspace…

LoRA Studio

Everything you need to train a great LoRA

A LoRA is a small add-on model that teaches AI one specific thing: your face, your pet, a product or an art style. In LoRA Studio you train it online from 10 to 30 pictures, compare checkpoints and create with it right away. No setup and nothing to install.

  • 10 to 30 pictures
  • Usually 15 to 60 minutes
  • 5 checkpoints to compare
  • Private, downloadable

What people train

People and characters

Put yourself or a fictional hero in any outfit, place and light.

Pets

Your dog or cat on the moon, in a painting or on a poster.

Art styles

Teach a look once, then create anything in it.

Products

Studio shots and lifestyle scenes without a photo shoot.

Build a dataset that trains well

  • Vary angles, light and backgrounds

    Different views, distances and settings teach the subject instead of one particular photo.

  • Use sharp, large pictures

    At least 768 pixels on the short side, in focus and without heavy filters.

  • One subject per picture

    Avoid group photos; crop to the person, pet or product you want to train.

  • Keep captions honest

    Captions start with your trigger word and are written for you. Fix anything that describes the picture wrongly.

Questions

What is a LoRA?

A LoRA (low-rank adaptation) is a small file that adds a new subject or style to an image model without retraining the whole model. You trigger it with a word in your prompt.

How many photos do I need?

Between 10 and 30. Quality and variety matter more than quantity: a dozen good, varied pictures beat 30 near-identical ones.

How long does training take?

Usually 15 to 60 minutes, depending on the number of pictures. Five checkpoints arrive with previews while it trains.

Which base model should I choose?

Krea 2 for photoreal and cinematic results with rich light; Z-Image Turbo for fast, versatile photos and illustration.

Can I download my LoRA?

Yes. Download any checkpoint whenever you want. Your LoRAs, datasets and images stay private to your account.

How much does it cost?

LoRA Studio is billed per hour while your session is open: training and creating are included. Unlimited members pay a lower hourly rate.

LoRA Studio is billed per hour while your session is open. Unlimited members pay a lower rate.

Where do you want to go?

↑↓ Navigate ↵ Open Esc Close

FLAT AI LORA STUDIO

Train a LoRA online: your own AI image model

Turn 10 to 30 photos into a custom AI model of a face, a pet, an art style or a product. LoRA Studio trains it on a dedicated GPU while you watch, then lets you create it in any scene, outfit and style.

How LoRA training works

  1. Add your pictures. Choose what the LoRA should learn and upload your photos. Each one is checked for quality and duplicates, and the captions are written for you.
  2. Train on Krea 2 or Z-Image Turbo. Pick the base model and how deep to train. Five checkpoints arrive while it learns, each with preview images, so you can see your model improve.
  3. Create and download. Keep the checkpoint you like best, write a prompt with your trigger word and create images in more than 20 styles, or download your LoRA as a .safetensors file.

What you can train

People and characters: your own face or a fictional hero, in new outfits, places and light. Pets: your dog or cat on a poster, in a painting or on the moon. Art styles: teach a look once and paint anything with it. Products: studio shots and lifestyle scenes without a photo shoot. Poses and concepts: a move, an outfit or an idea, on anyone.

Krea 2 or Z-Image Turbo

Krea 2 is photoreal and cinematic, with rich light and detail, and works with more than 20 styles. Z-Image Turbo is fast and versatile, for photos and illustration. Both learn from the same pictures, so you can train one and try the other later.

Pricing

LoRA Studio is billed by the hour while your studio session is open: 2.80 USD per hour, or 1.99 USD per hour with Unlimited. Training and creating images are included, and your LoRAs, pictures and images stay saved between sessions.

Frequently asked questions

What is a LoRA?

A LoRA is a small add-on file that teaches an image model something new, such as a face, a pet, a product or an art style, without retraining the whole model. You call it in a prompt with its trigger word.

How many photos do I need to train a LoRA?

At least 3, and 10 to 30 give the best results. Use sharp pictures of at least 768 pixels with different angles, light and backgrounds, and avoid filters and group photos.

How long does LoRA training take?

Usually 15 to 60 minutes, depending on how many pictures you add (up to 60) and the training depth you choose. Five checkpoints with previews arrive while it trains.

Which models can I train on?

Krea 2, photoreal and cinematic, and Z-Image Turbo, fast and versatile for photos and illustration.

Can I download my LoRA?

Yes. Every checkpoint can be downloaded as a .safetensors file.

How much does it cost?

LoRA Studio is billed by the hour while your studio session is open, from your Flat AI balance: 2.80 USD per hour, or 1.99 USD per hour with Unlimited. Training and creating images are included.

Private and responsible

Your pictures and LoRAs belong to your account and are not shared. Train only on people who agreed to it. Pictures of minors are not accepted.