Mergestein 角色写实模型是基于 NoobAI 架构深度训练的高精度角色生成大模型，专注于还原角色真实质感与细节表现力。模型在人体结构、皮肤纹理、服饰材质等维度实现精准建模，支持多样化角色风格适配，从写实人像到奇幻角色均能呈现自然光影过渡与细腻质感。
关于v2.0版本
我使用了第一个版本，将 CLIP 更换为更新的版本，并将其与我的 Uncanny Plus 版本稍作合并，这样一来，它的提示跟随性更好，对比度更低，色彩更自然，看起来也更逼真一些（逼真度和质量取决于使用的角色，有些角色看起来比其他角色更逼真、质量更好）。
不知何故，即使不使用胶片颗粒提示，它也能使图像看起来像是有某种压缩/颗粒效果，而且是同一个模型，只做了两处改动，所以我不知道它是怎么变成这样的，但看起来很有意思。
出现了同样的面部特效，这是 NoobAI 造成的，除非我移除它或者训练一个模型，否则无能为力。
手指有时看起来仍然会很奇怪，这可以通过增加否定提示来解决（在肯定提示中使用逼真的提示可以做到这一点，所以我建议测试一下使用和不使用这种提示的情况）。
建议设置
我上传的模型图片未经任何 LoRA、CFG Rescale、Hiresfix、ADetailer 或 Upscale 处理。因此，画质还有提升空间。和所有优秀的模型一样，近距离效果完美，中距离尚可，远距离则比较糟糕。
所有模型都已在模型中合并了 CFG 增强（在 CFG 较少的情况下效果很好，使用少于 5 个 CFG），因此不建议使用任何 CFG LoRa，否则只会得到烧毁的图像，并且 LoRa 还会影响一些细节增强。如果您仍然想使用它们，请使用较低的权重（小于 0.3）。
是的，它支持 NF/SW 内容（booru 标签），年龄限制也适用，所以请不要用于不良用途。
采样器、调度器和 CFG 建议：
Euler A Normal/simple/SGM Uniform 2.5 CFG (max 5 CFG) if you don't like skin texture (Skin pores, wrinkles, etc)Euler Beta 2.5 CFG (max 4.5 CFG) for better 3d/realism, close to DPM but with much less "noise".DPM++ 2/3/4M SDE Karras 2 CFG (Max 4 CFG, more the CFG = Brighter/burned). More realistic but too noisy and adds too many details. I only recommend using it on simple backgrounds and portraits/Upper bodyFor Euler/Euler A start with 16 steps max 30 steps.For DPM start with 20 steps max 35 steps.分辨率：
适用于所有 SDXL 分辨率，我个人使用 1024/1152×1536，有时也使用 1280×1536（这样可以获得身体扩展）。
推荐的基础提示：
积极的：
best quality, masterpiece, photorealistic, realistic,消极的：
worst quality, bad quality仅当您使用 A1111/Reforge 时才使用此方法，可提高质量：
[: worst quality, bad quality :0.2]提高真实感和质量的提示：
（这会降低字符提示的准确率，建议与 LoRas 或更高的权重一起使用）
Positive：
(best quality, masterpiece), realistic, photorealistic, photo (medium), [eye focus]Negative：
为了更好地理解角色提示并激发创造力，请移除 “(toon (style), 3d)”
worst quality, bad quality, (toon (style), 3D), traditional media, bad anatomy, bad proportions, bad hands, bad feet作品参照
best quality, masterpiece, cowboy shot, dutch angle, day, sunlight, outdoors, beach, coast, chromatic aberration, wind, BREAK
1girl, long hair, bangs, purple hair, blunt sidelocks, curly hair, medium breast, looking at viewer, denim shorts, yellow striped bikini top, yellow bow, yellow belt, light smile, shy, posing, <lora:Black_Desert_Online_StyleDetailer:0>
Negative prompt: bad quality, worst quality, bad hands, bad anatomy, bad proportions
Steps: 30, CFG scale: 3, Sampler: Euler, Seed: 1484261372, VAE: xlVAEC_f231.safetensors, Size: 1152x1536, Model: ComfyUI_00003_, Script: X/Y/Z plot, X Type: Prompt S/R, Version: f1.7.0-v1.10.1RC-latest-2190-g8731f1e9, X Values: [object Object], Pad conds: True, Model hash: cd1ac853b6, Schedule type: Beta, Beta schedule beta: 0.6, Beta schedule alpha: 0.6, Token merging ratio: 0.3
(best quality, masterpiece), realistic, night, dutch angle, light, party, upper body, from side, bokeh, bloom, soft focus, chromatic aberration, vignetting, film grain, BREAK
(vert (neptunia)), 1girl, cocktail dress, necklace, happy, closed mouth, looking at viewer, (posing), leaning on object, head rest
Negative prompt: worst quality, bad quality, bad hands, bad anatomy, bad proportions
Steps: 30, CFG scale: 3, Sampler: Euler, Seed: 232344139, VAE: xlVAEC_f231.safetensors, Size: 1152x1536, Model: ComfyUI_00002_, Version: f1.7.0-v1.10.1RC-latest-2190-g8731f1e9, Emphasis: No norm, Pad conds: True, Model hash: 93d39aebf6, Schedule type: Beta, Beta schedule beta: 0.6, Beta schedule alpha: 0.6, Token merging ratio: 0.3
lazyreal, lazympos, lazyup lazyup6, lazypos
cowboy shot Adult Female , Solo:1.2
gunmetal grey metallic foil pantyhose, liquid mercury chrome effect, reflective under light, 15 denier, detailed legs, sharp focus on legwear, high detail fabric texture, perfect anatomy|:1.2
Persian refined heart-shaped face, upturned cat-like eyes, high cheekbones, elegant aquiline nose |:1.2
cute dancer 20yo joyful smile rosy cheeks big eyes glossy lips ponytail ribbons blonde pink sequin crop frilly shorts leg warmers light-up sneakers twirl arms wide dance smoke beams|:1.2
No boots, no shoes,
eye-level, 70mm, shallow DOF,
(Close up:1.15), (cowboy shot:1.15),
(three-quarter view:1.2), (3/4 angle:1.2),
(subject in foreground:1.3), (dominant character:1.3),
<lora:For_Pony_v0.2:0.55>
<lora:add-detail-xl:0.88>
<lora:zy_Detailed_Backgrounds_v1:0.88>
Negative prompt: (camera:1.4)
ugly:1.2, BadX-neg DeepNegative_xl_v1 FastNegativeV2,( boots, shoes:1.1)
lazydn lazyhand lazyneg, censured, pain, blood,Watermark, Text, censored, deformed, bad anatomy, disfigured, poorly drawn face, mutated, extra limb, ugly, poorly drawn hands, missing limb, floating limbs, disconnected limbs, disconnected head, malformed hands, long neck, mutated hands and fingers, bad hands, missing fingers, cropped, worst quality, low quality, mutation, poorly drawn, huge calf, bad hands, fused hand, missing hand, disappearing arms, disappearing thigh, disappearing calf, disappearing legs, missing fingers, fused fingers, abnormal eye proportion, Abnormal hands, abnormal legs, abnormal feet, abnormal fingers
Steps: 48, CFG scale: 5, Sampler: DPM++ 2M SDE, Seed: 4089313302, TI: [object Object], Size: 1024x1536, Model: mergesteinUncanny_uncannyV2, Version: f2.0.1v1.10.1-previous-669-gdfdcbab6, Model hash: 93d39aebf6, Schedule type: Karras
Mergestein 角色写实模型是基于 NoobAI 架构深度训练的高精度角色生成大模型，专注于还原角色真实质感与细节表现力。模型在人体结构、皮肤纹理、服饰材质等维度实现精准建模，支持多样化角色风格适配，从写实人像到奇幻角色均能呈现自然光影过渡与细腻质感。
关于A Plus F版本
提高了中远距离的画质。
眼睛、手（指甲有时看起来还是有点奇怪）和 NAI 知识（danbooru 标签）略有改善。
建议设置
我上传的模型图片未经任何 LoRA、CFG Rescale、Hiresfix、ADetailer 或 Upscale 处理。因此，画质还有提升空间。和所有优秀的模型一样，近距离效果完美，中距离尚可，远距离则比较糟糕。
所有模型都已在模型中合并了 CFG 增强（在 CFG 较少的情况下效果很好，使用少于 5 个 CFG），因此不建议使用任何 CFG LoRa，否则只会得到烧毁的图像，并且 LoRa 还会影响一些细节增强。如果您仍然想使用它们，请使用较低的权重（小于 0.3）。
是的，它支持 NF/SW 内容（booru 标签），年龄限制也适用，所以请不要用于不良用途。
采样器、调度器和 CFG 建议：
Euler A Normal/simple/SGM Uniform 2.5 CFG (max 5 CFG) if you don't like skin texture (Skin pores, wrinkles, etc)Euler Beta 2.5 CFG (max 4.5 CFG) for better 3d/realism, close to DPM but with much less "noise".DPM++ 2/3/4M SDE Karras 2 CFG (Max 4 CFG, more the CFG = Brighter/burned). More realistic but too noisy and adds too many details. I only recommend using it on simple backgrounds and portraits/Upper bodyFor Euler/Euler A start with 16 steps max 30 steps.For DPM start with 20 steps max 35 steps.分辨率：
适用于所有 SDXL 分辨率，我个人使用 1024/1152x1536，有时也使用 1280x1536（这样可以获得身体扩展）。
推荐的基础提示：
积极的：
best quality, masterpiece, photorealistic, realistic,消极的：
worst quality, bad quality仅当您使用 A1111/Reforge 时才使用此方法，可提高质量：
[: worst quality, bad quality :0.2]提高真实感和质量的提示：
（这会降低字符提示的准确率，建议与 LoRas 或更高的权重一起使用）
Positive：
(best quality, masterpiece), realistic, photorealistic, photo (medium), [eye focus]Negative：
为了更好地理解角色提示并激发创造力，请移除 "(toon (style), 3d)"
worst quality, bad quality, (toon (style), 3D), traditional media, bad anatomy, bad proportions, bad hands, bad feet作品参照
best quality, masterpiece, realistic, photorealistic, cowboy shot, abstract background, gradient background, dutch angle, wind, ice, fantasy, light particles, chromatic aberration, bokeh, BREAK
(ganyu (genshin impact):1), 1girl, genshin impact, asian, sitting, posing, light smile, parted lips, detached sleeves, sideways glance
Negative prompt: [: worst quality, bad quality :0.2]
Steps: 30, CFG scale: 2.5, Sampler: DPM++ 4M SDE, Seed: 3190674076, Size: 1024x1536, Model: Uncanny-A PlusfixNAI, Version: f1.7.0-v1.10.1RC-latest-2190-g8731f1e9, Emphasis: No norm, Pad conds: True, Model hash: 22c1f9f342, Schedule type: Karras, Token merging ratio: 0.3best quality, masterpiece, realistic, photorealistic, feet out of frame, dutch angle, from above, white background, simple background, BREAK
(lucoa (maidragon):1), 1girl, kobayashi-san chi no maidragon, dragon girl, green eyes, santa dress, santa hat, capelet, shushing, seductive smile, looking at viewer, fur trim, posing, thighhighs
Negative prompt: [: extra digits, worst quality, bad quality :0.2]
Steps: 30, CFG scale: 2.5, Sampler: Euler, Seed: 575219674, Size: 1032x1536, Model: Uncanny-A PlusfixNAI, Version: f1.7.0-v1.10.1RC-latest-2190-g8731f1e9, Emphasis: No norm, Pad conds: True, Model hash: 22c1f9f342, Schedule type: Beta, Beta schedule beta: 0.6, Beta schedule alpha: 0.6, Token merging ratio: 0.3
图片提示词
此隐藏内容仅限VIP查看升级VIP
Mergestein 角色写实模型是基于 NoobAI 架构深度训练的高精度角色生成大模型，专注于还原角色真实质感与细节表现力。模型在人体结构、皮肤纹理、服饰材质等维度实现精准建模，支持多样化角色风格适配，从写实人像到奇幻角色均能呈现自然光影过渡与细腻质感。
这是我的Wai x Janku + NovaMoon-PM ILL XL的Mergestein测试模型，也是一个 Realistic Personal Merge，所以很多合并合并在一起，形成了一个合并。
NoobAi模型基于EPS版本。
这些图像未经任何 LoRA、CFG Rescale、Hiresfix、ADetailer 或 Upscale 处理，因此画质还有提升空间。和所有优秀的模型一样，近距离拍摄效果完美，中距离尚可，远距离则效果不佳。
是的，它支持 NF/SW 内容（booru 标签），年龄限制也适用，所以请不要用于不良用途。（如果您不想发布任何 NF/SW 内容，只需在正面提示中添加（通用）即可）
采样器、调度器和 CFG 建议：
Euler A Normal/simple/SGM Uniform CFG 3+ if you don't like skin texture (Skin pores, wrinkles, etc)Euler Beta CFG 2.5+ for better 3d/realism, close to DPM but with much less "noise".DPM++ 2/3/4M SDE Karras CFG 2.5+ More realistic but too noisy and adds too much skin details.For Euler/Euler A start with 16 steps max 30 steps.For DPM start with 20 steps max 35 steps.Resolution：
Works with all SDXL resolutions, i personally use 1024/1152x1536 and sometimes 1280x1536 (with this you can get body extension).推荐的基础提示：
积极的：
best quality, masterpiece, photorealistic, realistic,消极的：
（如果要减少皮肤纹理，请在否定词中添加以下提示：skin pores, skin texture）
[: hands, artistic error :0.2]作品参照
best quality, masterpiece, (realistic), (dutch angle), from above, upper body, BREAK
1girl, ((yunyun (konosuba):1), Kono Subarashii Sekai ni Shukufuku wo!, shy, blush, light smile, thighhighs, looking at viewer, hands between legs, sitting, (posing)), BREAK
atmospheric haze, (wind), flower, petals, nature
Negative prompt: [: hands, worst quality, bad quality, Low quality, worst aesthetic, lowres :0.2]
Steps: 35, CFG scale: 2.5, Sampler: Euler, Seed: 1904663715, Size: 1024x1536, Model: DURxWJPM+NvPM-S_Uncanny, Version: f1.7.0-v1.10.1RC-latest-2190-g8731f1e9, Emphasis: No norm, Pad conds: True, Model hash: 198ac6165d, Schedule type: Beta, Beta schedule beta: 0.6, Beta schedule alpha: 0.6, Token merging ratio: 0.3
best quality, masterpiece, (realistic), (dutch angle), (cowboy shot), from above, from side, BREAK
1girl, ((Aqua (konosuba):1), Kono Subarashii Sekai ni Shukufuku wo!, light smile, looking at viewer, white thighhighs, (blue thigh boots), sitting, hugging own legs, head rest, (posing:1)), BREAK
atmospheric haze, (wind), flower, petals, nature, lake
Negative prompt: [: hands, worst quality, bad quality, Low quality, worst aesthetic, lowres :0.2]
Steps: 30, CFG scale: 2.5, Sampler: Euler, Seed: 2437632373, Size: 1024x1536, Model: DURxWJPM+NvPM-S_Uncanny, Version: f1.7.0-v1.10.1RC-latest-2190-g8731f1e9, Emphasis: No norm, Pad conds: True, Model hash: 198ac6165d, Schedule type: Beta, Beta schedule beta: 0.6, Beta schedule alpha: 0.6, Token merging ratio: 0.3, rescale_cfg_enabled: True, rescale_cfg_version: Normal, rescale_cfg_multiplier: 0.4
best quality, masterpiece, realistic, cowboy shot, Dutch angle, general, from below, BREAK
1girl, ((sonozaki shion:1), higurashi no naku koro ni, seductive smile, hair ribbon, yellow jewelry, sleeveless sweater, green eyes, light smile, sitting, crossed legs, posing, pencil skirt, looking at viewer), BREAK
atmospheric haze, house, east asian architecture, japan, wind, indoors
Negative prompt: [: hands, worst quality, bad quality, Low quality, worst aesthetic, lowres, 3d :0.2]
Steps: 30, CFG scale: 2.5, Sampler: Euler, Seed: 663639198, Size: 1024x1536, Model: Mergestein-Uncanny, Version: f1.7.0-v1.10.1RC-latest-2190-g8731f1e9, Emphasis: No norm, Pad conds: True, Model hash: 198ac6165d, Schedule type: Beta, Beta schedule beta: 0.6, Beta schedule alpha: 0.6, Token merging ratio: 0.3 本内容及资源均来自网络，版权归原作者所有，仅供个人学习研究，请勿商用和侵犯他人肖像，如需商用请联系原作者。若原作者内容有侵权之处请速联系我们，我们将会在24小时内删除。
