Thaqib-Text2Image-3DPixarStyle

FLUX-Pixar-3D-Merged 🎬

This model is a fine-tuned version of black-forest-labs/FLUX.1-dev using LoRA adapters prithivMLmods/Canopus-Pixar-3D-FluxDev-LoRA to generate 3D images in a Pixar-like style. It was merged using fuse_lora() to produce a standalone, fully self-contained model ready for inference without requiring separate LoRA weights.

Model Details

Model Description

• Developed by: x-hayush
• Base model: black-forest-labs/FLUX.1-dev
• Model type: Diffusion-based image generation (FLUX pipeline)
• License: CreativeML Open RAIL++ (same as base model unless stated otherwise)
• LoRA source: prithivMLmods/Canopus-Pixar-3D-FluxDev-LoRA
• Merged with method: pipe.fuse_lora()

This model is designed to generate 3D characters, scenes, and illustrations in vibrant Pixar-like visual style using text prompts.

Uses

Direct Use

This model can be used directly to generate Pixar-style 3D images from text prompts. It's particularly well-suited for creative projects, illustrations, children's books, and visual storytelling.

from diffusers import FluxPipeline
import torch

pipe = FluxPipeline.from_pretrained(
    "x-hayush/FLUX-Pixar-3D-Merged",
    torch_dtype=torch.float16
).to("cuda")

image = pipe("a Pixar-style 3D child playing in a sunny garden").images[0]
image.show()

Out-of-Scope Use

• Not suitable for realistic photography or photo restoration
• Not designed for scientific/medical/biometric image generation

Bias, Risks, and Limitations

• As with most generative models, it may reflect visual and cultural biases from training data.
• May not generalize well to rare concepts or highly specific scenes without additional finetuning.

Training Details

• Finetuned with: LoRA adapter for style transformation
• LoRA merging method: fuse_lora() from Hugging Face diffusers API
• Precision: float16
• Hardware used: Google Colab with A100 GPU

Evaluation

Visual inspection of generations confirms improved fidelity in 3D Pixar-like outputs compared to base model. Formal metrics (CLIP-score, FID) not yet reported.

Environmental Impact

• Training Compute: Low (LoRA tuning + merge only)
• Hardware: Google Colab Pro (A100)
• Estimated Hours: <1h
• Carbon Emitted: Negligible

Model Card Contact

For questions or feedback, contact @x-hayush.