PenZlTr512 Raw Pencil &amp; Charcoal Sketch StudioA high-ranking artistic LoRA, trained to capture not just the texture of a pencil, but the actual gesture, pressure, dynamics, and searching lines of a living artist. The model completely removes the "neural network plasticity" and transforms generations into honest analog graphics with deep materiality.⚙️ Technical SpecificationsDataset: 200 high-quality, carefully selected and cleaned images (strict data synchronicity without domain shifts).Training Resolution: 1024x1024 pixels.Network Rank (Dim): 512 — extreme learning depth. Allows the model to memorize not pixel patterns, but stroke style, anatomical logic, and hand movement patterns.Precision: Trained without compromise and quantization to preserve pure mathematical gradients and subtle halftones.🚀 Recommended SettingsThe model responds sensitively to weight changes, dividing the graphics into two fundamentally different styles:Weight 0.60 – 0.75 (Soft Graphite / Academic Drawing): The deepest and most artistic mode. Produces the texture of a soft pencil, fine shading, neat crosshatching, and vibrant volume. Ideal for portraits, complex media, and natural objects.Weight 1.10 – 1.30 (Aggressive Ink / Liner / Lithography): Maximum contrast and pressure. The stroke becomes dense, like black paste or ink wash. Perfect for expressive concept art, geometry, and tattoo sketches.CFG Scale: Recommended to be kept around 4.5. This allows the model to breathe, maintains soft transitions, and doesn't burn black fractions into blind spots.Model Feature (Signature Artifact): Due to deep learning at rank 512, the model has fully adopted the concept of the artist's signature. Depending on the prompt and weight, it can independently integrate unique handwritten signatures into a drawing using the 512 or PenZlTr512 pen, delicately incorporating them into the image's plane and perspective.<a target="_blank" rel="ugc" href="https://github.com/OrakulStudio/AI-Toolkit-Viking-Engine-Fork">https://github.com/OrakulStudio/AI-Toolkit-Viking-Engine-Fork</a>

Z-image | Raw Pencil & Charcoal Sketch Studio | Rank 512

PenZITr512_000002000.safetensors

v1.0

Raw Pencil & Charcoal Sketch Studio

A high-ranking artistic LoRA, trained to capture not just the texture of a pencil, but the actual gesture, pressure, dynamics, and searching lines of a living artist. The model completely removes the "neural network plasticity" and transforms generations into honest analog graphics with deep materiality.

⚙️ Technical Specifications

Dataset: 200 high-quality, carefully selected and cleaned images (strict data synchronicity without domain shifts).

Training Resolution: 1024x1024 pixels.

Network Rank (Dim): 512 — extreme learning depth. Allows the model to memorize not pixel patterns, but stroke style, anatomical logic, and hand movement patterns.

Precision: Trained without compromise and quantization to preserve pure mathematical gradients and subtle halftones.

🚀 Recommended Settings

The model responds sensitively to weight changes, dividing the graphics into two fundamentally different styles:

Weight 0.60 – 0.75 (Soft Graphite / Academic Drawing): The deepest and most artistic mode. Produces the texture of a soft pencil, fine shading, neat crosshatching, and vibrant volume. Ideal for portraits, complex media, and natural objects.

Weight 1.10 – 1.30 (Aggressive Ink / Liner / Lithography): Maximum contrast and pressure. The stroke becomes dense, like black paste or ink wash. Perfect for expressive concept art, geometry, and tattoo sketches.

CFG Scale: Recommended to be kept around 4.5. This allows the model to breathe, maintains soft transitions, and doesn't burn black fractions into blind spots.

Model Feature (Signature Artifact): Due to deep learning at rank 512, the model has fully adopted the concept of the artist's signature. Depending on the prompt and weight, it can independently integrate unique handwritten signatures into a drawing using the 512 or PenZlTr512 pen, delicately incorporating them into the image's plane and perspective.

Files with SHA256: 10e61cfcab16c421f8c6c51a6c8fc6778cafa9138dbaeb18c622516f523df9ec