LoRACharacter and style training

Teach a modelyour look.

Upload your photos, we train the LoRA, you download it. Characters or styles.

Start training from $3.99 How it works

Pay only when training starts · Private 7-day workspace

Train on
Z-Image
Krea 2
Qwen-Image 2.1
MOSS-Image

How it works

Three steps.
Zero setup.

01
Upload
Create a workspace and add up to 100 character photos or 500 style images to your private link.
02
Train
Pay when you’re ready, then follow live status, checkpoints, and sample images as it trains.
03
Download
Grab your LoRA and use it in compatible Z-Image, Krea 2, Qwen-Image 2.1, or MOSS-Image workflows.

Training sets

Good photos in.
Great LoRA out.

Variety teaches the model what stays the same. This is what a strong set looks like.

Character set

Close-ups, full-body shots, new outfits, and new places. One face throughout.

Example character A: Close-up — Close-up

Example character A: Mirror selfie — Mirror selfie

Example character A: Golden hour — Golden hour

Example character B: Original — Original

Sunset profile

Example character B: Full body — Full body

Example character C: Original — Original

Example character C: Close-up — Close-up

Example character C: Boardwalk — Boardwalk

Style set

Up to 500 varied images that share one look. Style-focused captions are written for you.

Quick checklist

Mix close-ups, waist-up, and full-body shots
Vary pose, lighting, clothing, and background
Skip duplicates, blur, heavy filters, and sunglasses

Pricing

Pay per training.
Not per month.

Price follows your photo count. You see the exact total after upload, before you pay.

Photos30

Training

Model

Standard recipe 20% cheaper —

Max, highest settings—

With the recommended training steps. For character LoRAs the quality difference is usually small.

Train with these settings

FAQ

Good to know.

What is a LoRA?

A small add-on model that teaches an image model one specific character or visual style. Load it in a compatible workflow, use your trigger word, and generate new images of that character or style.

How is the price calculated?

By the number of photos you upload and the training steps. You upload first and see your exact total before paying. Removing duplicate or near-duplicate photos lowers the price, usually without reducing quality.

What should I upload?

Character training accepts up to 100 images. Mix close-ups, waist-up, full-body, and farther-away photos with different poses, lighting, clothing, expressions, angles, and backgrounds. Choose clear photos where the subject is easy to identify.

Avoid duplicates, blurry images, heavy filters, face-obscuring sunglasses, and unclear group shots. Style training accepts up to 500 varied images that share the look you want.

Should I choose Standard or Max?

Standard trains a very high quality LoRA (rank 64, or rank 128 on Krea 2) and costs 20% less. Max trains at the highest settings with full-rank LoKr. For character LoRAs, the quality difference is usually small.

Which models can I train on?

Z-Image, Krea 2, Qwen-Image 2.1, and MOSS-Image (SFT or Base).

Do I have to write captions?

No. AI captions are written for you, starting with your trigger word and class. Choose manual captions if you'd rather supply your own.

How long are my files kept?

Seven days. Training uploads, checkpoints, samples, and download files are deleted automatically after that, so download anything you want to keep.

Can I pay with Discord points?

Yes. Start from the /train-lora command in Discord to pay with points or by card. The exact point cost appears after upload.

Something else?