## 🎞️ Krea2 35mm Film Texture &amp; Cinematic Realism LoRA for Qwen-Image 2.1This LoRA brings **authentic 35mm motion-picture film texture, natural skin pores, tactile fabric weaves, and organic highlight/shadow roll-off** to **Qwen-Image 2.1**, while keeping **100% of Qwen-Image 2.1's native capabilities** (Text-to-Image, single &amp; multi-image Edit / I2I, character consistency, and text rendering).Vanilla Qwen-Image 2.1 has top-tier prompt adherence and editing power, but often produces overly smooth plastic skin, synthetic surface sheen, or a digital yellow-green cast. This LoRA removes that artificial digital look and replaces it with grounded, photorealistic cinema film texture — **with zero trigger words required (100% Plug &amp; Play).**---### ✨ Key Features* **Plug &amp; Play (No Trigger Word Needed)**: Trained directly across all 384 diffusion transformer blocks. Simply load the LoRA and write your normal prompts.* **Zero Loss of Native Capabilities**: Works seamlessly with both **T2I** and **Multi-Reference Image Editing <code>TextEncodeQwenImage21</code>)** without breaking character identity or prompt adherence.* **Natural Skin &amp; Surface Micro-Texture**: Restores realistic skin pores, fine hair strands, wool/tweed/velvet fabric weaves, leather grain, and stone/wood surfaces.* **Organic 35mm Film Color &amp; Lighting Roll-off**: Eliminates the flat digital/plastic look, adding rich shadow depth, natural halation, and fine 35mm film grain.* **Trained at Native 1920×1088 (16:9 FHD)**: Trained on **2,951 VLM-audited cinematic stills** at native <code>1920×1088</code> resolution (zero resizing or cropping) with captions 100% aligned to the official Qwen-Image 2.1 prompting structure.---### ⚙️ Recommended Settings| Parameter | Recommended Value | Notes || :--- | :--- | :--- || **Trigger Word** | **None Required** | Just load the LoRA and use your normal prompt || **LoRA Strength (I2I / Edit)** | *<code>0.50</code>** | Best balance for Character Sheets &amp; Multi-Image I2I (locks 100% facial identity) || **LoRA Strength (T2I)** | *<code>0.50 – 0.65</code>** | <code>0.50</code> for subtle natural realism, <code>0.60–0.65</code> for stronger 35mm film mood || **Base Model** | <code>Qwen-Image 2.1</code> | Fully compatible with both <code>bf16</code> and <code>int8_convrot</code> weights || **ComfyUI Placement** | <code>LoraLoaderModelOnly</code> | Connect right after *<code>Load Diffusion Model</code>** (before <code>QwenImage21Cache</code> / <code>ModelSamplingFlux</code>) || **Sampler / Scheduler** | <code>euler</code> / <code>simple</code> | Official Qwen-Image 2.1 flow-matching schedule || **Steps / CFG** | <code>40 steps</code> / <code>CFG 2.0</code> | Use <code>ModelSamplingFlux</code> <code>max_shift: 0.693548</code>, <code>base_shift: 0.5</code>) || **Resolution** | <code>1920×1088</code> / <code>2752×1536</code> (2K) | Works across all aspect ratios <code>16:9</code>, <code>8:9</code>, <code>3:4</code>, <code>9:16</code>, <code>1:1</code>) |---### 🧪 Training Specs* **Dataset**: 2,951 curated cinematic &amp; editorial frames at native <code>1920×1088</code> (filtered from 3,000 generations via automated VLM quality audit)* **Architecture**: Rank 64 <code>linear: 64</code>, <code>linear_alpha: 32</code>), <code>bf16</code>, <code>timestep_bias: low_noise</code>, targeting all 384 transformer LoRA modules (including <code>img_mlp</code>)---### 🇰🇷 한국어 안내 (Korean Guide)**Qwen-Image 2.1**의 모든 본래 기능(T2I 생성, 단일·다중 참조 이미지 편집/합성, 캐릭터 일관성, 텍스트 묘사력)을 그대로 유지하면서, 순정 모델 특유의 매끈한 플라스틱 피부결과 디지털 색감을 **실제 35mm 영화 필름 카메라의 피부 모공·원단 질감·깊은 암부 계조와 필름 입자감**으로 바꿔주는 LoRA입니다.* **트리거 워드**: **불필요** (별도 트리거 워드 없이 LoRA 연결만으로 즉시 작동합니다)* **추천 강도 <code>strength_model</code>)**: * *<code>0.50</code> (기본 권장 / I2I 캐릭터 시트·다중 이미지 편집용)**: 원본 얼굴 인상을 100% 그대로 유지하면서 자연스러운 피부결과 필름 톤 적용 * *<code>0.55 ~ 0.65</code> (T2I 단독 화보·영화 스틸컷용)**: 한층 진한 35mm 필름 질감과 시네마틱 무드 연출* **ComfyUI 연결 위치**: *<code>Load Diffusion Model</code> (디퓨전 모델 로드)** → *<code>LoraLoaderModelOnly (0.50)</code>** → <code>QwenImage21Cache</code> → <code>ModelSamplingFlux</code>

Files with SHA256: b94be18fa55f8389bd65236690fe03f8609b006e2cc428bfda7235836b94ded8