How to Train a Model ID

What is a Model ID?

A Model ID represents a model’s identity, similar to a fashion model’s name. The system uses LoRA (Low-Rank Adaptation) technology, sometimes called a LoRA ID. Users define AI fashion model attributes, generate training images reflecting those characteristics, then use these to train the Model ID for style mode generation.

Step 1: Enter Model ID Section and Click Create Button

Navigate to Model ID from the left sidebar and select “Create new model ID” to begin training.

Step 2: Select Training Images

Option 1: Upload Existing Images

Upload training images directly via the right panel, or select previously saved images from Photo Gallery by filtering and clicking selections.

Option 2: Create Training Images from Scratch

A. Generate Reference Face

Click “Generate reference face” on the Create New Images tab to customize facial parameters.

Users can generate up to 4 face options for comparison and select “Set as Reference Face.”

B. Upload Reference Image

Users may directly upload existing faces to the reference box. Important: Real person faces require explicit permission for training purposes.

Step 3: Customize Attributes and Pose Prompts

After obtaining a reference image, customize model attributes by deselecting unwanted parameters to preserve reference image defaults.

Selected parameters auto-populate the prompt template. Choose poses, camera angles, and set quantity per category before generating training data.

Edit prompts directly or uncheck pose categories as needed. Submit generation requests to create consistent-face model images, then select qualifying images and click “Save Selected” to move selections rightward. Repeat until sufficient training data accumulates.

Step 4: Start Training

Enter the LoRA model ID name and specify whether training uses real human or AI-generated models. Real human image training requires model permission. Status displays “Training” during processing, changing to “Trained” upon completion.

Pro Tips: Training Image Selection

Approximately 20 training images ensure LoRA model ID stability. Recommended distribution: one-third full-body, one-third half-body, one-third close-up shots. Include front-facing, back-facing, and profile views across various angles. Include pose examples if specific positioning is required.

Step 5: Test Model ID

Access the Model Generation Tool to validate the trained Model ID. Upload or select reference images, choose similarity settings (more/less similar to reference), select your trained LoRA model, generate prompts based on reference images, then generate outputs.

Submitted generation requests appear as cards in the History section. Click refresh to monitor job status. Upon completion, review generation results and preview images at enlarged sizes to confirm successful training and expectation alignment. Retrain if unsatisfied.

Step 6: Manage Model ID

Trained Model IDs automatically appear on the Model ID page.

Delete Model ID: Click the select button (upper right), check desired deletions, click delete. Deletion cannot be undone—use cautiously.

Rename Model ID: Click the pencil icon adjacent to Model ID. Changes auto-save upon focus loss.

Important Note: No duplicate prevention exists for Model IDs currently. Avoid creating identically-named Model IDs to prevent confusion.

FAQs

What is Virtual Try-On?

HuHu AI’s virtual try-on tool creates realistic garment visualizations on customizable virtual models using advanced AI, delivering efficient, affordable fashion content with creative control.

Are the virtual try on photos licensed for commercial use?

Yes. HuHu provides copyright-free preset AI models with diverse skin tones, ages, and features usable for commercial purposes. Users must ensure uploaded images don’t infringe third-party rights and verify local legal compliance.

What are the advantages of AI virtual try-on compared to traditional photoshoots?

AI virtual try-on significantly reduces production costs, delivers results within minutes, enables full attribute control (pose, body type, styling), and promotes sustainability by eliminating physical samples, travel, and studio waste.

How can I create my own AI models for virtual try on?

Users can fully customize AI fashion models via the model generator feature, selecting body shape, skin tone, hairstyle, facial features, or uploading reference photos for similar-look AI generation.