tyche-labs/lekhaAI_flux_v1

🪶 LekhaAI Flux v1

An AI model built to craft elegant and stylish invitation template images, blending creativity with sophistication for any occasion.

🧠 Model Details

Property	Description
Developed by	Tyche Labs
Model Type	Diffusion
Base Model	black-forest-labs/FLUX.1-dev
Language	English (for text prompts)
License	MIT
Tags	text-to-image, invitations, diffusion, design, LoRA, FLUX

📘 Description

LekhaAI Flux v1 is a fine-tuned variant of the FLUX 1-Dev diffusion model, optimized to generate elegant, aesthetic, and text-aware invitation template designs.
It excels at balancing artistic composition and layout structure, making it ideal for:

• 💍 Weddings
• 🎉 Birthdays
• 🏢 Corporate Events
• 🌸 Anniversaries and Milestones

🎨 Training Details

Attribute	Description
Base Model	black-forest-labs/FLUX.1-dev
Technique	LoRA fine-tuning
Resolution	Custom
Optimizer	AdamW
Dataset Size	~210 prompt–image pairs
Training Duration	~12 hours on NVIDIA A100
Loss Function	Standard diffusion loss
LoRA Rank	16

💡 Intended Use

✅ Primary Uses

• Generate invitation templates for various events.
• Provide creative starting points for designers.
• Serve as a base for further fine-tuning.

⚠️ Limitations

• Not for photo-realistic human faces.
• Not optimized for >1024 px resolutions.
• Occasional text/artifact blending possible in title areas.

⚙️ Python Code

from diffusers import FluxPipeline
import torch

# Hugging Face access token (required for base model)
HF_TOKEN = "hf_your_access_token_here"

# Base model and LoRA weight paths
MODEL_NAME = "black-forest-labs/FLUX.1-dev"
LORA_PATH = "./lekhaAI_lora_flux.safetensors"  # local or absolute path

# Load the base FLUX model
pipe = FluxPipeline.from_pretrained(
    MODEL_NAME,
    torch_dtype=torch.bfloat16,
    token=HF_TOKEN
).to("cuda")

# Load your fine-tuned LoRA weights
pipe.load_lora_weights(LORA_PATH)

# (optional) fuse LoRA for slight speedup at inference time
pipe.fuse_lora(lora_scale=0.8)

# Generate an image
prompt = "Elegant watercolor floral wedding invitation with gold geometric frame"

image = pipe(
    prompt=prompt,
    num_inference_steps=30,
    guidance_scale=7.5,
    width=768,
    height=1024
).images[0]

# Save the output
image.save("invitation_flux_lora.png")

Model generated images with descriptions

A cheerful brunch invitation with a soft blue background, framed by hand-painted orange branches, lush green leaves, and coral blossoms. Playful peach dots add a lively touch, creating a fresh and inviting citrus theme. A festive Indian invitation design featuring an ornate golden arch, vibrant floral patterns, glowing diyas, and a graceful peacock. The pastel background and rich saffron, maroon, and royal blue tones create an elegant, celebratory feel. A pastel watercolor invitation with wildflowers, dreamcatchers, fairy lights, and rustic wooden arches. Add butterflies and soft greenery along the edges, leaving the center blank for text. Use blush pink, sage green, cream, and gold tones for a bohemian, elegant vibe. A watercolor-style tropical invitation featuring palm leaves, coconuts, hibiscus flowers, and pineapples. Add playful toucans and parrots perched on branches. Use a vibrant palette of teal, coral, sunshine yellow, and green. Keep the middle blank for party details. A vibrant space-themed invitation with a deep navy background, surrounded by colorful planets, stars, and constellations in shades of purple, teal, pink, and yellow. The center is left blank for text, creating a fun, dreamy cosmic vibe. A cozy Christmas invitation featuring a decorated tree, gifts, and balloons on a snowy base. Holly garlands and red ornaments frame the border, with confetti adding a festive touch in warm red, green, and gold tones.

🧑‍🔬 Authors & Contributors

• Harihara Yashwanth Veldanda
• Harsha Chintamaneni
• Mrunal Kulakarni