MdxHeadingH1

MdxParagraph

MdxHeadingH2

MdxHeadingH3

StepGuide

Use 8–25 high-quality images (12 recommended) showing different angles, expressions, and lighting, while keeping the character's core design consistent. Crop or resize all images to square format—512x512 or 768x768 is ideal. Avoid logos, watermarks, or unowned content.

Prepare your character images

Access the Freepik AI Image Generator. Select "Character", then "New character". Name your LoRA and upload your images.

Upload your dataset

Use the default settings unless you need specific adjustments. Add a unique token or keyword (e.g., `leo-char`) for prompting. Start training—depending on image size and server load, this takes 10–30 minutes.

Set your training options

Embed

What is the Freepik AI Assistant?


<MdxHeadingH1>Custom characters</MdxHeadingH1>

<MdxParagraph>With a [LoRA character](https://www.freepik.com/ai/custom-characters), you can train the AI to understand and recreate your design with reliable accuracy. This guide shows you how to create consistent characters with AI using LoRA training.</MdxParagraph>

<MdxHeadingH2>What is a custom character?</MdxHeadingH2>

<MdxParagraph>A custom character is a unique figure, human, creature, or stylized avatar that stays visually consistent across multiple AI generations.</MdxParagraph>

<MdxParagraph>LoRA (Low-Rank Adaptation) technology helps you teach the AI how your character looks. Once trained, the AI recognizes your custom character and can **recreate it in different scenes, poses, or styles**.</MdxParagraph>

<MdxHeadingH2>How to create consistent characters with AI</MdxHeadingH2>

<MdxParagraph>To create your character more accurately, use photos from different angles with varied expressions, lighting, and backgrounds.</MdxParagraph>

<MdxHeadingH3>Add key traits to reference images</MdxHeadingH3>

<MdxParagraph>Include clear visual traits in your reference images, such as pink hair or a smile. This helps the AI recognize and keep those features.</MdxParagraph>

<MdxHeadingH3>Limit visual noise</MdxHeadingH3>

<MdxParagraph>Overly detailed images can interfere with character accuracy. If the results look off, simplify the image to focus on key traits.</MdxParagraph>

<MdxHeadingH2>How to train a character LoRA</MdxHeadingH2>

<StepGuide steps={[
  {
    title: 'Prepare your character images',
    description: 'Use 8–25 high-quality images (12 recommended) showing different angles, expressions, and lighting, while keeping the character\'s core design consistent. Crop or resize all images to square format—512x512 or 768x768 is ideal. Avoid logos, watermarks, or unowned content.'
  },
  {
    title: 'Upload your dataset',
    description: 'Access the Freepik AI Image Generator. Select "Character", then "New character". Name your LoRA and upload your images.'
  },
  {
    title: 'Set your training options',
    description: 'Use the default settings unless you need specific adjustments. Add a unique token or keyword (e.g., `leo-char`) for prompting. Start training—depending on image size and server load, this takes 10–30 minutes.'
  },
]} />

<Embed youtubeId="jYBkQVnPgQc" title="What is the Freepik AI Assistant?" allowPause />

<MdxHeadingH2>Credit use and rights</MdxHeadingH2>

<MdxParagraph>Training a LoRA and generating a character costs **5,500 credits per character**.</MdxParagraph>

<MdxParagraph>All generated content follows [Freepik's AI Terms and Conditions](https://www.freepik.com/legal/terms-of-use). Don't include copyrighted or trademarked material in your training images without proper rights.</MdxParagraph>

<MdxParagraph>Once your LoRA character is ready, continue with [Objects](https://www.freepik.com/ai/docs/objects).</MdxParagraph>