🌿 Craftly-Image-Turbo (12.8B)
🎨 Showcase Gallery
| 🌸 Anime Girl in Nature | 🌿 4K Macro Mossy Forest | 🏛️ Minimalist Travertine Pavilion |
|---|---|---|
(Explore the full 50-image 4K showcase gallery in the 📁 images/ directory).
🚀 One-Click Interactive Studios (Colab, Kaggle, Marimo)
You can launch and generate images immediately using any of our ready-to-run studios in the notebooks/ directory:
1. 🟡 Google Colab Studio
- Notebook:
notebooks/CRAFTLY_GOOGLE_COLAB.ipynb - Features: Interactive UI sliders, curated prompt presets (Anime Girl, 4K Nature, Minimalist Zen, Cyberpunk City), and automatic image download to your computer.
- Hardware: Runs smoothly on free T4 GPU, V100, L4, or A100.
2. 🔵 Kaggle Studio
- Notebook:
notebooks/CRAFTLY_KAGGLE.ipynb - Features: Optimized for Kaggle Dual T4 or P100 GPUs, automatic VRAM clearing, batch image generation loop, and output to
/kaggle/working/.
3. 🟣 Marimo Reactive Web App
- Script:
notebooks/CRAFTLY_MARIMO.py - Notebook:
notebooks/CRAFTLY_MARIMO_JUPYTER.ipynb - Features: Full reactive browser app with
marimo run. Real-time text prompt area, step sliders, resolution selector, and instant reactive image generation.
💻 Python Quickstart
import torch
from diffusers import DiffusionPipeline
REPO_ID = "CraftlyrobotMushfiqur/testy"
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16 if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else torch.float16
# 1. Load sovereign pipeline
pipe = DiffusionPipeline.from_pretrained(
REPO_ID,
custom_pipeline="pipeline_craftly",
torch_dtype=dtype,
trust_remote_code=True
).to(device)
# 2. Ultra-Fast 8-Step Synthesis
prompt = (
"A breathtaking masterpiece of a beautiful anime girl standing in a sunlit lush green forest meadow, "
"gentle breeze fluttering her long silky hair and delicate white dress, glowing cherry blossoms and emerald leaves "
"floating in the air, crystal clear sparkling stream beside her, Makoto Shinkai and Studio Ghibli aesthetic, "
"cinematic volumetric golden hour sunbeams, radiant expressive eyes, peaceful smile, vibrant 4k anime wallpaper, hyper-detailed"
)
with torch.inference_mode():
image = pipe(
prompt=prompt,
num_inference_steps=8,
guidance_scale=0.0,
width=1024,
height=1024
).images[0]
image.save("craftly_output.png", quality=95)
print("✅ Saved to craftly_output.png")
🏗️ Architecture Specifications
- Pipeline Class:
CraftlyImagePipeline - Transformer Diffusion Backbone:
CraftlyImageTransformer2DModel(28 Layers, 48 Attention Heads, 64-channel Latent Input) - Text Conditioning Backbone:
CraftlyVisionModel(36 Layers, 2560 Hidden Dim) - Latent Autoencoder: 16-channel Spatial VAE
- Denoising Trajectory: Flow-Matching Euler Discrete Scheduler (Optimized 8-step Distillation)
- Native Resolution: 1024×1024 pixels
⚖️ Citation & Attribution
@article{craftlyimageturbo2026,
title={Craftly-Image-Turbo: High-Velocity Visual Generation via Distilled Flow-Matching},
author={Md Mushfiqur Rahim and Craftly AI Research Team},
year={2026},
publisher={Hugging Face}
}