如果觉得生成效果的真实感还不够，可以尝试搭配一些 Real / Realism 类的 LoRA，通常能进一步提升画面的写实感和质感。If you feel the results are not realistic enough, you can try adding some Real / Realism LoRAs. They can usually help improve the overall realism and texture of the images.V3 UpdateV3 增加了更多女性角色的多样性，并进一步优化了皮肤质感，让整体表现更加自然、细腻。V3 adds more diversity to female characters and further improves skin texture, resulting in a more natural and refined overall look.V2：噪点更低、更平滑，但细节可能略少于 V1.5。新版本不代表一定更好，请按需选择。V2: Less noise and smoother results, but slightly less detail than V1.5. Newer doesn’t always mean better—choose based on your preference.<h2 id="1.5">1.5</h2>增加了一些真实感，新增int8、bf16|Improved realism; added INT8 and BF16.<h2>中文说明</h2>VelvetMuse 是一款基于 Krea 2 Turbo 的 FP8 模型，主要面向写实成人角色、人像摄影、皮肤与服饰质感、丝袜细节、姿势表现及整体画面美化。模型采用 Turbo 架构，能够以较少的采样步数实现快速生成，适合写实人像、时尚摄影和成人题材创作。<h3>模型信息</h3><ul><li>基础模型：Krea 2 Turbo</li><li>精度：FP8 Scaled</li><li>格式：Safetensors</li><li>触发词：无</li><li>成人内容：是</li><li>文件大小：约 12.24 GiB</li></ul><h3>推荐设置</h3><ul><li>Steps：8</li><li>CFG：1.0</li><li>Sampler：Euler</li><li>Scheduler：Simple</li><li>建议分辨率：1024 级别或更高</li><li>推荐使用自然语言描述人物、服装、姿势、镜头、构图和光线</li></ul><h3>模型特点</h3><ul><li>写实成人角色与摄影质感</li><li>细腻的皮肤和服饰表现</li><li>增强丝袜材质与腿部细节</li><li>更丰富的人物姿势</li><li>较好的光影、色彩和整体画面表现</li><li>Turbo 快速生成</li></ul><h2 id="english-description">English Description</h2>VelvetMuse is an FP8 model based on Krea 2 Turbo, designed for realistic adult characters, portrait photography, refined skin and clothing textures, hosiery details, expressive poses, and polished visual quality.Built on the Turbo architecture, it delivers fast generation with a low step count and is suitable for realistic portraits, fashion photography, and adult-oriented artwork.<h3 id="model-information">Model Information</h3><ul><li>Base Model: Krea 2 Turbo</li><li>Precision: FP8 Scaled</li><li>Format: Safetensors</li><li>Trigger Word: None</li><li>Adult Content: Yes</li><li>File Size: Approximately 12.24 GiB</li></ul><h3 id="recommended-settings">Recommended Settings</h3><ul><li>Steps: 8</li><li>CFG: 1.0</li><li>Sampler: Euler</li><li>Scheduler: Simple</li><li>Recommended Resolution: 1024-class or higher</li><li>Detailed natural-language prompts are recommended for subjects, clothing, poses, camera angles, composition, and lighting</li></ul><h3 id="features">Features</h3><ul><li>Realistic adult character rendering</li><li>Refined skin and clothing textures</li><li>Enhanced hosiery and leg details</li><li>More expressive character poses</li><li>Improved lighting, color, and overall visual quality</li><li>Fast Turbo generation</li></ul>

VelvetMuse_Krea2_Turbo

velvetmuseKrea2Turbo_v30_bf16.safetensors

velvetmuseKrea2Turbo_v30_nvfp4.safetensors

velvetmuseKrea2Turbo_v30_int8.safetensors

Files with SHA256: c44465d086790e75d82fc147d4fba7d34ef6025d335b087090318779fc87b60c