Sign In
homemodelsimagesvideospostsarticleschallengeseventsupdatesshop

The AI model that finally knows your thing.

Train a custom AI model on your own characters, styles, and concepts, then generate them on demand inside Civitai. Consistent every time. Training starts around 50 cents, with a 30-day free window on every model you train.

Train a Model
Eight portraits of the same character produced by a stock AI model, visibly drifting in face, outfit, and art style.
Stock model, drifting every render
Eight consistent, on-model portraits of the same character produced by a custom-trained AI model on Civitai.
Your trained model, same character every time

Over 50,000 AI models trained on Civitai every month.

Why stock AI models miss, and why training your own fixes it.

You've been sketching the same character for a year. You have a style people recognized before you had a name for it. You have a concept nobody else has trained. Stock models don't know any of that. They give you a close cousin. They give you a guess.

You write a 400-word prompt to get the character right. You regenerate nine times to get one usable frame. You finally get the face right, but the jacket changes. You fix the jacket and the style evaporates. You open a commercial image-editor, pay per edit, and still end up with a jawline that drifts every third generation.

The problem isn't your prompt.The problem is that the model has never seen the thing you're actually trying to make.

What is a fine-tuned AI model (LoRA)?

Big open-source image models have seen millions of paintings, photographs, characters, and scenes. But they've never seen your character, your palette, or the jacket you designed.

A fine-tuned model solves that. You take a base model that already understands images in general, and you show it a small, focused set of examples, usually 15 to 30. Training nudges the model until it can reliably reproduce whatever those examples have in common. What you're left with is a LoRA, a lightweight add-on that plugs into the base model and teaches it one new trick.

A general model learns general things.A fine-tuned model learns your thing.

Think of it as a generator that's read your notes.

Diagram showing how fine-tuning works: a base model plus your example images produces a fine-tuned model.

What training your own model unlocks.

Nine consistent portraits of the same character, showing perfect character consistency across lighting, pose, and expression.
Character consistency that doesn't drift
Same jaw, same jacket, same scar, from key art to the fifteenth panel. No more “close enough” across seeds, sessions, or months.
The same landscape scene rendered in four distinct artistic styles, demonstrating fine-tuned style capture.
A signature style, not a preset
A line weight, a palette, a grain, the look people clock as yours. Train it once and stop describing it every prompt.
An invented mascot creature rendered consistently across four different scenarios, demonstrating a custom concept trained into an AI model.
Custom concepts stock models can't produce
A prop you invented, a creature from your worldbuilding, a mascot, a uniform, a piece of architecture. Fifteen good references can teach the model something the internet has never trained on.
Cost comparison between per-edit commercial AI tools and a flat-rate custom-trained model, where the per-edit curve climbs while the trained-model line stays flat.
Predictable, affordable costs
Commercial image-editors bill per edit, and the total climbs every iteration. On Civitai, training starts at ~50¢ and every generation afterward runs at the regular generator rate. Generate 30 times as many images for the cost of a single edit on other platforms.

How training works on Civitai.

Three steps. No GPU to rent, no notebook to babysit, no LoRA files to hand-manage.

1
Upload your examples
Drop in 15 to 30 images (or video clips) of your subject. Our trainer reads each one and writes a short caption describing what's in it. Captions are the text labels the model uses to learn what each image contains. Auto works for most jobs; you can edit captions when you want more precise control.
2
Pick an engine
Civitai supports the full range of open-source base models, with new engines landing as they ship. You can even fine-tune on top of another fine-tune, stack your training on top of a community favorite and go deeper.
3
Generate onsite
When training finishes, your model lands in your Civitai library. Load it into the generator like any other model and start prompting.
Start Training

Training starts at 500 Buzz (~50¢). Once it's done, your trained model is free to use for 30 days.

Free for 30 days after you train.

Train today, generate with it for 30 days, no subscription required. Keep it after that with a Civitai membership.

Day 0
Training complete
Your freshly trained model lands in your Civitai library.
Day 1 to 30
Use it free
Generate with it inside the Civitai generator, no subscription needed. Try it on real work.
After Day 30
Keep it or publish it
Keep it private with a membership, or publish it to the community for free. Your call.

Keep your model private.

You curated the dataset. You designed the character. You own the IP. A private model stays on your account, out of the public library, and nobody else can load, scrape, or data-mine it. A Civitai membership is how you keep it past the 30-day window. The tier decides how many private models you can have in rotation.

Bronze
3

Private models

A character, a world, a style. Start here.
Silver
10

Private models

A full project. A cast. A universe.
Gold
100

Private models

A studio's worth. Series, franchises, clients, experiments, all live at once.
Compare PlansTrain for ~50¢ first

Private-mode generation onsite is limited to PG and PG-13.

Frequently asked questions

Make the thing only you can make.

The generator already knows a million styles. None of them are yours yet. Train the one that is.

Train Your First Model