README.md

---
tags:
- text-to-image
- lora
- diffusers
- template:diffusion-lora
widget:
- output:
    url: images/hammer_output.png
  text: >-
    HNDPNT, Texture the wireframe mesh of a complex ornate fantasy tech-hammer.
    Apply a stylized handpainted base color style. The texture features a blocky
    head constructed from polished gold trim, copper plating, and silver inlays
    with scrollwork. It is embedded with large faceted glowing blue gems on the
    sides and blue crystal spikes at the back. The handle displays copper
    piping, vertical blue energy vials, and a dark brown grip with glowing blue
    runes, ending in a mechanical claw pommel.
- output:
    url: images/barrel_output.png
  text: >-
    HNDPNT, Texture the wireframe mesh of a stylized wooden barrel. Apply a
    stylized handpainted base color style. The texture features a body
    constructed from thick, dark brown weathered wooden staves with prominent
    stylized wood grain, deep gaps, and edge wear with lighter brown highlights.
    It is reinforced by three broad, dark grey hammered metal bands secured with
    large, chunky metal rivets. The top surface is a circular wooden lid with a
    central recessed metal handle ring. Thick brushstrokes and baked-in lighting
    define the chunky forms.
base_model: nunchaku-tech/nunchaku-qwen-image-edit-2509
instance_prompt: HNDPNT
license: apache-2.0
---
# Wireframe to Texture LoRA

<Gallery />

## Model description 

# Wireframe to Hand-Painted – Stylized Texture LoRA

This LoRA is designed to convert **3D wireframe projections into stylized, hand-painted base textures** using image-to-image generation.  
It is intended to be used as a **base texture generator**, not as a final production asset tool.

Check out this post for more info, prompts examples, etc: https:&#x2F;&#x2F;www.artstation.com&#x2F;artwork&#x2F;vb6X9a

---

## 🔹 Base Model

This LoRA was trained on the **base version of Qwen Image Edit**  
⚠️ **NOT trained on Qwen Image Edit 2509**  
Results and behavior are aligned with the **base model only**.

---

## 🔹 Training Setup

- Trained on **ModelScope**
- Dataset size: **50+ curated image pairs**
- Training data:
  - **Input:** 2D image of Wireframe &#x2F; projection renders from 3D models  
  - **Output:** Stylized hand-painted texture results
- Multiple LoRA variations were trained and tested
- Each variation included **multiple checkpoints**
- The released version corresponds to the **best-performing checkpoint** after visual and consistency evaluation

---

## 🔹 Intended Input

This model expects:

- A **clean wireframe or projection render** of a 3D model
- **Neutral background** (gray or flat color recommended)
- **Clear silhouette and readable geometry**

Optimized for:

- Stylized &#x2F; hand-painted textures
- Fantasy and stylized props
- Game-ready asset base generation

---

## ✅ What This Model Is For

- **Fast base texture generation**
- **Stylized look exploration**
- **Early production &amp; iteration**
- **Indie and small-team pipelines**
- **Artist-in-the-loop workflows**

---

## ❌ What This Model Is NOT

- ❌ Not a one-click final asset generator  
- ❌ Does **not** generate full PBR texture sets  
- ❌ Does **not** replace manual texturing  
- ❌ Only textures **visible projected faces** (no full 360° automatic coverage)

---

## 🔹 Recommended Workflow

1. Export a **wireframe &#x2F; projection render** from your 3D software (3DCoat, Blender, etc.)
2. Use **image-to-image** with this LoRA
3. Generate a **stylized base texture**
4. **Refine manually** in 3DCoat, Substance, or painting tools

---

## 🧠 Philosophy

This model is built to **empower artists, not replace them**.  
It removes the **blank canvas problem** and accelerates early-stage production, while keeping **full creative control** in the artist’s hands.

AI as a **starting point**, not the final answer.


## Trigger words

You should use `HNDPNT` to trigger the image generation.


## Download model


[Download](/ianlebot/Wireframe_to_Texture/tree/main) them in the Files & versions tab.