fal/flux-klein-9b-virtual-tryon-lora

Flux Klein 9b Virtual Try-On LoRA

Virtual clothing try-on LoRA for Flux Klein 9b Edit. Given a person image and clothing reference images, generates a realistic result of the person wearing the specified outfit.

3 input images (person, top, bottom) → 1 output with the person wearing the clothes.

Built with fal.ai

Try it now:

• Playground (direct) — upload your images and run the model
• Workflow with LLM — auto-generates the prompt for you

Results

Highlights

Feature	Details
Base Model	Flux Klein 9b Edit
Input	3 images: person + top garment + bottom garment
Output	Person wearing the specified outfit
Resolution	Up to 1024px
Clothing Types	T-shirts, hoodies, jackets, jeans, skirts, pants, shorts
Works With	Men and women, various body types and poses

Examples

Woman — Graphic Tee + Distressed Jeans

Woman — Varsity Jacket + Cargo Shorts

Man — NASA Tee + Japanese Joggers

Man — Flight Bomber Jacket + Patchwork Jeans

Man — Polaroid Camera Tee + Striped Trousers

Example images shown on this page were generated using Nano Banana Pro and are not part of the training data.

Prompt Format

TRYON [person description]. Replace the outfit with [top description] and [bottom description] as shown in the reference images. The final image is a full body shot.

Image Input Order

The model expects 3 reference images in this specific order:

Position	Image	Description
1st	Person / Mannequin	The person to dress
2nd	Top garment	T-shirt, hoodie, jacket, etc.
3rd	Bottom garment	Jeans, pants, skirt, shorts, etc.

Prompt Examples

TRYON young woman with long dark wavy hair, standing casually. Replace the outfit
with black oversized t-shirt with Japanese kanji and neon pink Tokyo skyline graphic
and heavily distressed ripped light blue denim jeans as shown in the reference images.
The final image is a full body shot.

TRYON young man with medium blonde wavy hair, standing casually. Replace the outfit
with black oversized t-shirt with holographic silver NASA space logo and black jogger
sweatpants with white Japanese characters as shown in the reference images.
The final image is a full body shot.

Weights

Two versions of the LoRA weights are provided:

File	Use with
flux-klein-tryon.safetensors	fal.ai API / diffusers
flux-klein-tryon-comfy.safetensors	ComfyUI

Recommended Settings

Parameter	Value
num_inference_steps	28
guidance_scale	2.5
lora_scale	1.0

Usage

Try It Online

The easiest way to try this model:

• Playground (direct) — upload 3 images and write the prompt yourself
• Workflow with LLM — upload 3 images and the LLM auto-generates the prompt

Python (fal-client)

import fal_client

result = fal_client.submit(
    "fal-ai/flux-2/klein/9b/base/edit/lora",
    arguments={
        "prompt": "TRYON young woman with long dark hair, standing casually. Replace the outfit with red hoodie and blue jeans as shown in the reference images. The final image is a full body shot.",
        "image_urls": [
            "https://your-person-image.png",    # 1st: person
            "https://your-top-image.png",        # 2nd: top garment
            "https://your-bottom-image.png",     # 3rd: bottom garment
        ],
        "loras": [{
            "path": "YOUR_LORA_WEIGHTS_URL",
            "scale": 1.0,
        }],
        "num_inference_steps": 28,
        "guidance_scale": 2.5,
    }
).get()

print(result["images"][0]["url"])

Tips for Best Results

• Describe the person accurately in the prompt (hair, pose, body type)
• Describe the clothing in detail (color, pattern, style, material)
• Full body shots work best — mention "full body shot" in the prompt
• Clean backgrounds on input images improve results
• Consistent lighting across the 3 input images helps
• The trigger word TRYON must be at the start of the prompt
• Keep lora_scale at 1.0 for best results
• Works with both photographic and AI-generated input images

Training

• Base Model: Flux Klein 9b Edit
• Steps: 2000
• Learning Rate: 5e-05
• Platform: fal.ai

Full Showcase

22 outfits across men and women with diverse clothing styles.

Author

Created by Lovis Odin at fal.ai