YiffyMix 动物怪物大模型，是一款极为独特且功能强大的创新型大模型。它深度融合了前沿的 NoobAI 技术，巧妙依托 sdxl 与 sd1.5 的底层架构，历经无数次算法优化与数据精细调校，得以构建出一个能够精准生成各类栩栩如生、形态各异动物怪物形象的智能创作平台 。
v6.1 [2D,3D,真实,分辨率：896~1536] ※NoobXL eps 版本（非 v-pred）。
更加逼真的细节，减少动漫风格，修复平淡无趣的背景。
使用e621 标签（无下划线），艺术家标签在 YiffyMix 中非常有效
Species(v60)/Artists(v51) list (!!update!!) & Furry LoRAs (SDXL)/samples/wildcards
为新用户推荐艺术家列表和舒适的 UI 工作流程
YiffyMix XL 不需要 Pony score tags.
不要在 SDXL 中使用 SD1.5 vae，否则会得到烧焦的图像。
示例设置
设置 SDXL (SDXL-lightning & NoobXL)
- Steps = 12~24
- Sampler = “DDPM”, “Euler A SGMUniform”, “Euler SGMUniform”
- CFG scale = 4
- Negative embeddings SDXL = ac-neg1, ac-neg2 (你实际上不需要这个)
- Postive LoRA SDXL = SeaArt Quality Tags LoRA
- Stop at CLIP layers = 2
设置SD 1.5 + vpred
- Steps = 30~40
- Sampler = “DPM++ 2M Karras”, “DPM++ SDE Karras”, “DDIM”, “UniPC”
- CFG scale = 6~8
- Negative embeddings = deformity_v6, bwu [SD-WebUI\embeddings]
- Stop at CLIP layers = 1
- Eta noise seed delta = 0
- SD VAE = kl-f8-anime2, Furception [SD-WebUI\models\VAE]
Hires. fix
- Hires steps = Steps * Denoising strength
- Denoising strength = 0.25
- Hires upscaler = 4x-UltraMix_Smooth [SD-WebUI\models\ESRGAN]
Img2img 第一遍 ( Upscale enhance )
- Scale to = 1.5
- Denoising strength = 0.4~0.45
- Extra ( seed ) = true
- Variation seed = -1 ( use same seed )
- Variation strength = 0.9
Img2img 第二遍 ( Detail enhance )
- Scale to = 1
- Denoising strength = 0.3~0.35
- Extra ( seed ) = true
- Variation seed = -1 ( use same seed )
- Variation strength = 0.95
ControlNet
- ControlNet = softedge_hed, control_v11p_sd15_softedge
- ControlNet SDXL = softedge_hed, sdxlsoftedge-dexined, noobai-xl-controlnet
- ControlNet Weight = 0.35~0.5
- ControlNet Pixel Perfect = true
SD WebUI
LoRA Training
- imgs count = 15~50
- total steps = epoch * imgs count * folder loop = 3000~4500
- network_dim = 64
- network_alpha = 128 (SDXL) \ 16 (Noob)
- learning_rate = 0.0002
- unet_lr = 0.0001
- text_encoder_lr = 0.00005
- lr_scheduler = “cosine_with_restarts”
- mixed_precision = “bf16”
- optimizer_type = “Adafactor”
- optimizer_args = [ “scale_parameter=False”, “relative_step=False”, “warmup_init=False”, ]
YiffyMix v4x V-pred 设置
# YiffyMix v4x A1111 SD-WebUI 或 Forge V-pred 设置
1.下载配置文件“.yaml”并粘贴到模型旁边。
2.Raname“ .yaml”与模型名称相同。（检查“ .yaml”是否有[parameterization：“v”]行）
3.重新启动 SD-WebUI（如果配置加载失败，模型只会产生噪音）
# YiffyMix v4x ComfyUI 或StableSwarmUI V-pred 设置
添加节点“ModelSamplingDiscrete”，Sampling = “v_prediction”
# ComfyUI V-pred 设置（旧版本）
1.将配置“.yaml”复制到[ComfyUI\models\configs]并刷新ComfyUI
2.“加载检查点（带配置）” [右键\AddNode\advanced\model_merging]
工作流程：[加载检查点（带配置）]-[KSampler]-[VAE 解码]-[保存图像]
# v-pred 模式故障排除
如果您使用新版本的 webui-forge 并且检测 v-pred 模型失败。
尝试更新 webui-forge（单击 update.bat）。
当加载 v-pred 模型时，您会在 cmd 控制台中看到此行：
剩余键：dict_keys([‘v_pred’])
# v4.x 有时会出现这种情况，图像会有点烧坏
使用“带有 __wildcards__ 提示的动态提示”和“批次大小 > 1”
只需使用带有参数的相同提示再次生成，它就会恢复正常。
版本信息
v4.4 [2D,3D,真实,512~1088,VAE烘焙]
v-pred 版本，不再有黄色/棕色问题。
该版本不需要 CFG Rescale
v5.0 [2D,3D,真实,896~1536]
SDXL-闪电版本，基于 Compassmix XL。
v5.1 [2D,3D,真实,896~1536]
SDXL-闪电版本，更多 e621 数据集，更低的人脸，更好的 NF/SW 内容。
v5.2 [2D,3D,真实,分辨率：896~1536]
SDXL-lightning 版本，提高平均质量。比 v51 稍不稳定，但更有创意。
v6.0 [2D,3D,半真实,分辨率：896~1536]
NoobXL eps 版本（不是 v-pred）。
更多角色更好的XA姿势，有限且不受控制的风格（主要是动漫风格）。
基本风格
负面提示 SD1.5
unusual anatomy, mutilated, malformed, watermark,
amputee, mosaic censorship, sketch, monochrome
负面提示 SDXL
malformed, worst quality, bad quality,
signature, text, url
3D 艺术品风格
Prompt:
blender \(software\), ray tracing, 3d, unreal engine
写实风格
Prompt:
[:photorealistic, analog style, realistic, photorealism:0.5]
Negative prompt:
[:bwu:0.5]
推荐Refiner:
IndigoFurryMix v110-Realistic Switch at 0.5
Prompt SDXL v5x:
by Mandy Disher, by Wim Wenders, by Robert Rauschenberg, [by Fossa666::0.65],
ultra realistic, photorealism, photograph,
Prompt SDXL v6x-Noob:
[3d \(artwork\):(ultra realistic, photorealism, photograph:1.35):0.35]
Negative prompt SDXL:
sketch, manga, vector, line art, toony
一些使用 a1111 [a:b:value] 脚本提示的 SD 技巧：
1.高细节增加
[:some style or artist:0.15]
Let first 10%~20% steps free, then style will create on high detail base.
2.避免人物姿势受到艺术家或背景的影响
[:artist or background prompt:0.2]
Skipping first 20% steps can avoid this.
3.”A artist character” + “B artist style”
[A artist:B artist:0.35]
Most base concept is done in first 30%~50% steps, last steps only change style.
作品参照
nick wilde, black cloak, fluffy, cyber armor, one hands holding light saber, angry, circle effect, reaching towards viewer, action pose, open mouth, half-length portrait, [high-angle view::0.1], BREAK, masterpiece, best quality, high quality, by Kenket, zootopia, [:realistic:0.3], detailed background, depth of field, ambient silhouette, backlighting, darkness, abandoned ruins sci-fi science lab
Negative prompt: cybernatic, police, malformed, worst quality, bad quality
Steps: 17, CFG scale: 4, Sampler: Euler A SGMUniform, Seed: 3355818894, Size: 960x1280, Model: YiffyMix v61, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Model hash: 90226f38ce, Clip skip: 2
solo neferpitou, shadow creature, claws, [crouching:reaching towards viewer:0.15], action pose, glowing eyes, leaning forward, three-quarter portrait, BREAK, masterpiece, best quality, high quality, horror \(theme\), by Syvaron, [:realistic:0.75], detailed background, ambient silhouette, depth of field, backlighting, [:dark night, darkness plant, speed line, ruins, storm:0.1], purple light, yellow light
Negative prompt: malformed, worst quality, bad quality, censor bar, signature, text, url
Steps: 17, CFG scale: 4, Sampler: Euler A SGMUniform, Seed: 1191903233, Size: 960x1280, Model: YiffyMix v61, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Model hash: 90226f38ce, Clip skip: 2
solo furry curvy figure, husky, big eyes, medium breasts, police uniform, police hat, high-angle view, looking at viewer, fluffy, three-quarter view, open smile, standing, leaning forward, particle effects, BREAK, masterpiece, best quality, high quality, [:realistic, photorealism, photograph:0.25], detailed background, depth of field, [:street, window, morning:0.1], light, plant
Negative prompt: english text, malformed, worst quality, bad quality
Steps: 17, CFG scale: 4, Sampler: Euler A SGMUniform, Seed: 3287065391, Size: 960x1280, Model: YiffyMix v61, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Model hash: 90226f38ce, Clip skip: 2
使用介绍
使用e621 标签（无下划线），艺术家标签在 YiffyMix 中非常有效
Species(v60)/Artists(v51) list (!!update!!) & Furry LoRAs (SDXL)/samples/wildcards
为新用户推荐艺术家列表和舒适的 UI 工作流程
YiffyMix XL 不需要 Pony score tags.
不要在 SDXL 中使用 SD1.5 vae，否则会得到烧焦的图像。
示例设置
设置 SDXL (SDXL-lightning & NoobXL)
- Steps = 12~24
- Sampler = “DDPM”, “Euler A SGMUniform”, “Euler SGMUniform”
- CFG scale = 4
- Negative embeddings SDXL = ac-neg1, ac-neg2 (你实际上不需要这个)
- Postive LoRA SDXL = SeaArt Quality Tags LoRA
- Stop at CLIP layers = 2
设置SD 1.5 + vpred
- Steps = 30~40
- Sampler = “DPM++ 2M Karras”, “DPM++ SDE Karras”, “DDIM”, “UniPC”
- CFG scale = 6~8
- Negative embeddings = deformity_v6, bwu [SD-WebUI\embeddings]
- Stop at CLIP layers = 1
- Eta noise seed delta = 0
- SD VAE = kl-f8-anime2, Furception [SD-WebUI\models\VAE]
Hires. fix
- Hires steps = Steps * Denoising strength
- Denoising strength = 0.25
- Hires upscaler = 4x-UltraMix_Smooth [SD-WebUI\models\ESRGAN]
Img2img 第一遍 ( Upscale enhance )
- Scale to = 1.5
- Denoising strength = 0.4~0.45
- Extra ( seed ) = true
- Variation seed = -1 ( use same seed )
- Variation strength = 0.9
Img2img 第二遍 ( Detail enhance )
- Scale to = 1
- Denoising strength = 0.3~0.35
- Extra ( seed ) = true
- Variation seed = -1 ( use same seed )
- Variation strength = 0.95
ControlNet
- ControlNet = softedge_hed, control_v11p_sd15_softedge
- ControlNet SDXL = softedge_hed, sdxlsoftedge-dexined, noobai-xl-controlnet
- ControlNet Weight = 0.35~0.5
- ControlNet Pixel Perfect = true
SD WebUI
LoRA Training
- imgs count = 15~50
- total steps = epoch * imgs count * folder loop = 3000~4500
- network_dim = 64
- network_alpha = 128 (SDXL) \ 16 (Noob)
- learning_rate = 0.0002
- unet_lr = 0.0001
- text_encoder_lr = 0.00005
- lr_scheduler = “cosine_with_restarts”
- mixed_precision = “bf16”
- optimizer_type = “Adafactor”
- optimizer_args = [ “scale_parameter=False”, “relative_step=False”, “warmup_init=False”, ]
YiffyMix v4x V-pred 设置
# YiffyMix v4x A1111 SD-WebUI 或 Forge V-pred 设置
1.下载配置文件“.yaml”并粘贴到模型旁边。
2.Raname“ .yaml”与模型名称相同。（检查“ .yaml”是否有[parameterization：“v”]行）
3.重新启动 SD-WebUI（如果配置加载失败，模型只会产生噪音）
# YiffyMix v4x ComfyUI 或StableSwarmUI V-pred 设置
添加节点“ModelSamplingDiscrete”，Sampling = “v_prediction”
# ComfyUI V-pred 设置（旧版本）
1.将配置“.yaml”复制到[ComfyUI\models\configs]并刷新ComfyUI
2.“加载检查点（带配置）” [右键\AddNode\advanced\model_merging]
工作流程：[加载检查点（带配置）]-[KSampler]-[VAE 解码]-[保存图像]
# v-pred 模式故障排除
如果您使用新版本的 webui-forge 并且检测 v-pred 模型失败。
尝试更新 webui-forge（单击 update.bat）。
当加载 v-pred 模型时，您会在 cmd 控制台中看到此行：
剩余键：dict_keys([‘v_pred’])
# v4.x 有时会出现这种情况，图像会有点烧坏
使用“带有 __wildcards__ 提示的动态提示”和“批次大小 > 1”
只需使用带有参数的相同提示再次生成，它就会恢复正常。
版本信息
v4.4 [2D,3D,真实,512~1088,VAE烘焙]
v-pred 版本，不再有黄色/棕色问题。
该版本不需要 CFG Rescale
v5.0 [2D,3D,真实,896~1536]
SDXL-闪电版本，基于 Compassmix XL。
v5.1 [2D,3D,真实,896~1536]
SDXL-闪电版本，更多 e621 数据集，更低的人脸，更好的 NF/SW 内容。
v5.2 [2D,3D,真实,分辨率：896~1536]
SDXL-lightning 版本，提高平均质量。比 v51 稍不稳定，但更有创意。
v6.0 [2D,3D,半真实,分辨率：896~1536]
NoobXL eps 版本（不是 v-pred）。
更多角色更好的XA姿势，有限且不受控制的风格（主要是动漫风格）。
基本风格
负面提示 SD1.5
unusual anatomy, mutilated, malformed, watermark,
amputee, mosaic censorship, sketch, monochrome
负面提示 SDXL
malformed, worst quality, bad quality,
signature, text, url
3D 艺术品风格
Prompt:
blender \(software\), ray tracing, 3d, unreal engine
写实风格
Prompt:
[:photorealistic, analog style, realistic, photorealism:0.5]
Negative prompt:
[:bwu:0.5]
推荐Refiner:
IndigoFurryMix v110-Realistic Switch at 0.5
Prompt SDXL v5x:
by Mandy Disher, by Wim Wenders, by Robert Rauschenberg, [by Fossa666::0.65],
ultra realistic, photorealism, photograph,
Prompt SDXL v6x-Noob:
[3d \(artwork\):(ultra realistic, photorealism, photograph:1.35):0.35]
Negative prompt SDXL:
sketch, manga, vector, line art, toony
一些使用 a1111 [a:b:value] 脚本提示的 SD 技巧：
1.高细节增加
[:some style or artist:0.15]
Let first 10%~20% steps free, then style will create on high detail base.
2.避免人物姿势受到艺术家或背景的影响
[:artist or background prompt:0.2]
Skipping first 20% steps can avoid this.
3.”A artist character” + “B artist style”
[A artist:B artist:0.35]
Most base concept is done in first 30%~50% steps, last steps only change style.
作品参照
nick wilde, cross necklace, priest, gold and black cloak, sunglasses, holding revolver, portrait, action pose, serious face, smoking, half-length portrait, [high-angle view::0.1], BREAK, masterpiece, best quality, [:by Duskihorns, by Youjomodoki:0.15], detailed background, depth of field, ambient silhouette, backlighting, night, arch, ruins, dungeon, plant, foggy, factory, snowing, grave
Negative prompt: window, malformed, worst quality, bad quality
Steps: 17, CFG scale: 4, Sampler: Euler A SGMUniform, Seed: 1720726479, Size: 960x1280, Model: YiffyMix v60, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Mask blur: 10, Model hash: 37d9e3df33, Denoising strength: 0, Clip skip: 2
furry miku, kemono fox, jojo pose, young, green hair, blue green body, black short skirt, black strapless crop top, black detached sleeves, cyan tie, BREAK, masterpiece, best quality, jojo pose, detailed background, ambient silhouette, stage, spot light, purple light, blue light, sparkles
Negative prompt: human, tan skin, malformed, worst quality, bad quality
Steps: 17, CFG scale: 4, Sampler: Euler A SGMUniform, Seed: 2684485685, Size: 960x1280, Model: YiffyMix v60, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Model hash: 37d9e3df33, Clip skip: 2
solo muscular lagiacrus, scalie, horn, claws, action pose, half-length portrait, looking back at viewer, three-quarter view, BREAK, masterpiece, best quality, monster hunter, [:by Dmitrys, by Paul Lehr:0.45], detailed background, ambient silhouette, depth of field, backlighting, underwater, ruins cathedral, glowing plant, crystal, purple light, yellow light, foggy, [:cliff, pillar, stained glass:0.1]
Negative prompt: breasts, female, malformed, worst quality, bad quality, signature, text, url
Steps: 17, CFG scale: 4, Sampler: Euler A SGMUniform, Seed: 2793471322, Size: 960x1280, Model: YiffyMix v60, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Model hash: 37d9e3df33, Clip skip: 2
使用介绍
关于v52
partA = Compassmix XL + Boltning v1.0 * 0.35
partB = (partA + BananaStrike XL – SDXL Lightning 4 step) * trainDiff:0.25
YiffyMix v52 = (partB + Compassmix XL – SDXL Lightning 4 step) * trainDiff:0.25
提醒
使用e621 标签（无下划线），艺术家标签在 YiffyMix 中非常有效。
物种/艺术家网格列表（更新 v44+v37）和Furry LoRA/样本/通配符
VAE：WD-KL-F8-Anime2，Furception 校正对比度和降噪 [Settings\Stable Diffusion\SD VAE]
ClipSkip：:Stop at CLIP layers=1 (SD=1、NAI=2、SDXL=2) [Settings\Stable Diffusion\Clip skip]
推荐新用户使用“艺术家列表” ，在 v3-v4 YiffyMix 版本中使用。
### YiffyMix v50 SDXL-lightning 设置 ###
YiffyMix v50 不需要 PonyXL score tags。
不要在 SDXL 中使用 SD1.5 vae，否则会得到烧焦的图像。
使用设置 [steps=12~24, cfg=4~6]
推荐采样器：
平滑：[“DDPM”, “Euler A Turbo”, “Euler A SGMUniform”]
细节和平衡：[“Euler”, “Euler SGMUniform”]
高细节和高对比度：[“DPM++ 2M SDE SGMUniform”]
# A1111 SD-WebUI 或 Forge V-pred 设置 #
1.下载配置文件“.yaml”并粘贴到模型旁边。
2.Raname“ .yaml”与模型名称相同。（检查“ .yaml”是否有[parameterization：“v”]行）
3.重新启动 SD-WebUI（如果配置加载失败，模型只会产生噪音）
# ComfyUI 或StableSwarmUI V-pred 设置 #
添加节点“ModelSamplingDiscrete”，Sampling = “v_prediction” [右键单击\AddNode\advanced\model]
工作流程：[加载CheckPoint]-[ModelSamplingDiscrete]-[KSampler]-[VAE解码]-[保存图像]
# ComfyUI V-pred 设置（旧版）#
1.将配置“.yaml”复制到[ComfyUI\models\configs]并刷新ComfyUI
2.“加载检查点（带配置）” [右键\AddNode\advanced\model_merging]
工作流程：[加载检查点（带配置）]-[KSampler]-[VAE 解码]-[保存图像]
# 一些故障排除 #
如果您使用新版本的 webui-forge 并且检测 v-pred 模型失败。
尝试更新 webui-forge（单击 update.bat）。
当加载 v-pred 模型时，您会在 cmd 控制台中看到此行：
剩余键：dict_keys([‘v_pred’])
# v4.x 有时会出现这种情况，图像会有点烧坏问题#
使用“带有 __wildcards__ 提示的动态提示”和“批次大小 > 1”
只需使用带有参数的相同提示再次生成，它就会恢复正常。
示例设置
Txt2img Setting
- Steps = 30~40
- Sampler = DPM++ 2M Karras, DPM++ SDE Karras, DDIM, UniPC
- CFG scale = 6~8
- Negative embeddings = deformity_v6, bwu ※[SD-WebUI\embeddings]
- Negative embeddings SDXL = ac-neg1, ac-neg2※[SD-WebUI\embeddings]
Hires. fix
- Hires steps = 15
- Denoising strength = 0.45~0.5
- Hires upscaler = 4x-UltraMix_Smooth ※[SD-WebUI\models\ESRGAN]
Img2img Resize
- Scale to = 1.5~2.0
- Denoising strength = 0.45~0.5
- ControlNet0 = softedge_hed, control_v11p_sd15_softedge
- ControlNet0 Weight = 0.4~0.5
- ControlNet0 Pixel Perfect = true
- ControlNet1 = Reference
- ControlNet1 Weight = 0.2~0.3
- ControlNet1 Pixel Perfect = true
SD Setting
- Stop at CLIP layers = 1
- Eta noise seed delta = 0
- SD VAE = kl-f8-anime2 ※[SD-WebUI\models\VAE]
Txt2img Animated Diff ※no “BREAK”, tokens < 75
- Steps = 30~50
- Sampler = DDIM
- CFG = 10~20
- ImageSize = 512~768 ※large image less motion
- model = mm-Stabilized_mid, motionModel_v03-anime
- Number of frames = 10~14
负面提示 v2
unusual anatomy, pixel \(artwork\), mutilated, malformed, watermark, amputee,
restricted palette, mosaic censorship, deformity, sketch, unfinished, monochrome
3D艺术风格
Prompt:
masterpiece, best quality, light, 4k, 2k,
blender \(software\), ray tracing, 3d, unreal engine
写实风格
Prompt:
[:photorealistic, analog style, realistic, photorealism:0.5]
Negative prompt:
[:bwu:0.5]
Recommend Refiner:
IndigoFurryMix v110-Realistic
Switch at 0.5
| 使用提示： | CLIP SKIP: 2 |
作品参照
head up, looking_up, half-length portrait, standing, male, nick wilde, sunglasses, hand above head, aloha shirt, smile, teeth, golden necklace, BREAK, vaporwave, subsurface scattering, by Conrad Roset, by M.W. Kaluta, detailed background, detailed foreground, depth of field, ambient silhouette, backlighting, sunset, cityscape, night, planet, beach, plam, cloud, (sport car:0.75), ambient butterfly
Negative prompt: female, eyeshadow, 3d \(artwork\), malformed, worst quality, bad quality
Steps: 17, CFG scale: 4, Sampler: DDPM, Seed: 2020395195, Size: 1024x1280, Model: YiffyMix v52, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Mask blur: 10, Model hash: 293a85966b, Denoising strength: 0
quadruped feral glaceon, excessive fluff, wizard hat, high collar cloak, lying on front, glowing eyes, BREAK, [by Cotora, by Aoizuri:realistic, photorealism:0.85], detailed background, detailed foreground, depth of field, ambient silhouette, backlighting, inside, night, ambient butterfly, flower, castle, book, plant, waterfall
Negative prompt: malformed, worst quality, bad quality
Steps: 17, CFG scale: 4, Sampler: DDPM, Seed: 90245881, Size: 1024x1280, Model: YiffyMix v52, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Model hash: 293a85966b
three-quarter portrait, solo, howl, head up, red sun, quadruped amaterasu \(okami\), fluffy tail, yellow sclera, (wolf:0.5), leaning forward, looking away, BREAK, ink, watercolor \(artwork\), monochrome, by Jean-Paul Riopelle, by Katsushika Hokusai, detailed background, detailed foreground, glowing red light, black sky, cloud forest, mountain, japanese mythology, (huge wave:0.65)
Negative prompt: eyes closed, toony, chibi, malformed, worst quality, bad quality
Steps: 17, CFG scale: 4, Sampler: DDPM, Seed: 823577908, Size: 1024x1280, Model: YiffyMix v52, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Model hash: 293a85966b
使用介绍
关于v51
YiffyMix v51 = (YiffyMix v50 + Compassmix XL – SDXL Lightning 2 step) * trainDiff:0.55
提醒
使用e621 标签（无下划线），艺术家标签在 YiffyMix 中非常有效。
物种/艺术家网格列表（更新 v44+v37）和Furry LoRA/样本/通配符
VAE：WD-KL-F8-Anime2，Furception 校正对比度和降噪 [Settings\Stable Diffusion\SD VAE]
ClipSkip：:Stop at CLIP layers=1 (SD=1、NAI=2、SDXL=2) [Settings\Stable Diffusion\Clip skip]
推荐新用户使用“艺术家列表” ，在 v3-v4 YiffyMix 版本中使用。
### YiffyMix v50 SDXL-lightning 设置 ###
YiffyMix v50 不需要 PonyXL score tags。
不要在 SDXL 中使用 SD1.5 vae，否则会得到烧焦的图像。
使用设置 [steps=12~24, cfg=4~6]
推荐采样器：
平滑：[“DDPM”, “Euler A Turbo”, “Euler A SGMUniform”]
细节和平衡：[“Euler”, “Euler SGMUniform”]
高细节和高对比度：[“DPM++ 2M SDE SGMUniform”]
# A1111 SD-WebUI 或 Forge V-pred 设置 #
1.下载配置文件“.yaml”并粘贴到模型旁边。
2.Raname“ .yaml”与模型名称相同。（检查“ .yaml”是否有[parameterization：“v”]行）
3.重新启动 SD-WebUI（如果配置加载失败，模型只会产生噪音）
# ComfyUI 或StableSwarmUI V-pred 设置 #
添加节点“ModelSamplingDiscrete”，Sampling = “v_prediction” [右键单击\AddNode\advanced\model]
工作流程：[加载CheckPoint]-[ModelSamplingDiscrete]-[KSampler]-[VAE解码]-[保存图像]
# ComfyUI V-pred 设置（旧版）#
1.将配置“.yaml”复制到[ComfyUI\models\configs]并刷新ComfyUI
2.“加载检查点（带配置）” [右键\AddNode\advanced\model_merging]
工作流程：[加载检查点（带配置）]-[KSampler]-[VAE 解码]-[保存图像]
# 一些故障排除 #
如果您使用新版本的 webui-forge 并且检测 v-pred 模型失败。
尝试更新 webui-forge（单击 update.bat）。
当加载 v-pred 模型时，您会在 cmd 控制台中看到此行：
剩余键：dict_keys([‘v_pred’])
# v4.x 有时会出现这种情况，图像会有点烧坏问题#
使用“带有 __wildcards__ 提示的动态提示”和“批次大小 > 1”
只需使用带有参数的相同提示再次生成，它就会恢复正常。
示例设置
Txt2img Setting
- Steps = 30~40
- Sampler = DPM++ 2M Karras, DPM++ SDE Karras, DDIM, UniPC
- CFG scale = 6~8
- Negative embeddings = deformity_v6, bwu ※[SD-WebUI\embeddings]
- Negative embeddings SDXL = ac-neg1, ac-neg2※[SD-WebUI\embeddings]
Hires. fix
- Hires steps = 15
- Denoising strength = 0.45~0.5
- Hires upscaler = 4x-UltraMix_Smooth ※[SD-WebUI\models\ESRGAN]
Img2img Resize
- Scale to = 1.5~2.0
- Denoising strength = 0.45~0.5
- ControlNet0 = softedge_hed, control_v11p_sd15_softedge
- ControlNet0 Weight = 0.4~0.5
- ControlNet0 Pixel Perfect = true
- ControlNet1 = Reference
- ControlNet1 Weight = 0.2~0.3
- ControlNet1 Pixel Perfect = true
SD Setting
- Stop at CLIP layers = 1
- Eta noise seed delta = 0
- SD VAE = kl-f8-anime2 ※[SD-WebUI\models\VAE]
Txt2img Animated Diff ※no “BREAK”, tokens < 75
- Steps = 30~50
- Sampler = DDIM
- CFG = 10~20
- ImageSize = 512~768 ※large image less motion
- model = mm-Stabilized_mid, motionModel_v03-anime
- Number of frames = 10~14
负面提示 v2
unusual anatomy, pixel \(artwork\), mutilated, malformed, watermark, amputee,
restricted palette, mosaic censorship, deformity, sketch, unfinished, monochrome
3D艺术风格
Prompt:
masterpiece, best quality, light, 4k, 2k,
blender \(software\), ray tracing, 3d, unreal engine
写实风格
Prompt:
[:photorealistic, analog style, realistic, photorealism:0.5]
Negative prompt:
[:bwu:0.5]
Recommend Refiner:
IndigoFurryMix v110-Realistic
Switch at 0.5
| 使用提示： | CLIP SKIP: 2 |
作品参照
solo nick wilde, bedroom eyes, sunglasses, straw hat, aloha shirt, green eyes, portrait, selfie, smile, teeth, v sign, BREAK, by Danji-Isthmus, by Dangpa, by M.W. Kaluta, island, town, summer day, beach, plant, water
Negative prompt: malformed, worst quality, bad quality
Steps: 17, Size: 896x1280, Seed: 353768045, Model: YiffyMix v51, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: Euler SGMUniform, CFG scale: 4, Model hash: 90c8c9e41f
solo feral green yoshi tyrannosaurus rex, yellow sclera, claws, sharp teeth three-quarter portrait, three-quarter view, looking at viewer, standing, BREAK, by Agnes Cecile, by Paul Lehr, jurassic park, detailed background, detailed foreground, depth of field, ambient silhouette, backlighting, movie poster, museum, jungle, abandoned building, jurassic park banner
Negative prompt: size difference, quadruped, malformed, worst quality, bad quality
Steps: 17, Size: 896x1280, Seed: 898430425, Model: YiffyMix v51, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: DDPM, CFG scale: 4, Model hash: 90c8c9e41f
pikachu, deerstalker hat, fluffy fur, full-length portrait, big eyes, slightly chubby BREAK, unreal engine, pokmon detective pikachu, detailed background, detailed foreground, depth of field, ambient silhouette, backlighting, cityscape, restaurant, street, plant, neon light, purple light
Negative prompt: malformed, worst quality, bad quality, anime style
Steps: 17, Size: 896x1280, Seed: 3973518573, Model: YiffyMix v51, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: DDPM, CFG scale: 4, Model hash: 90c8c9e41f
使用介绍
使用e621 标签（无下划线），艺术家标签在 YiffyMix 中非常有效。
物种/艺术家网格列表（更新 v44+v37）和Furry LoRA/样本/通配符
VAE：WD-KL-F8-Anime2，Furception 校正对比度和降噪 [Settings\Stable Diffusion\SD VAE]
ClipSkip：:Stop at CLIP layers=1 (SD=1、NAI=2、SDXL=2) [Settings\Stable Diffusion\Clip skip]
推荐新用户使用“艺术家列表” ，在 v3-v4 YiffyMix 版本中使用。
### YiffyMix v50 SDXL-闪电设置 ###
YiffyMix v50 不需要 PonyXL score tags。
不要在 SDXL 中使用 SD1.5 vae，否则会得到烧焦的图像。
使用设置 [steps=12~24, cfg=4~6]
推荐采样器：
平滑：[“DDPM”, “Euler A Turbo”, “Euler A SGMUniform”]
细节和平衡：[“Euler”, “Euler SGMUniform”]
高细节和高对比度：[“DPM++ 2M SDE SGMUniform”]
# A1111 SD-WebUI 或 Forge V-pred 设置 #
1.下载配置文件“.yaml”并粘贴到模型旁边。
2.Raname“ .yaml”与模型名称相同。（检查“ .yaml”是否有[parameterization：“v”]行）
3.重新启动 SD-WebUI（如果配置加载失败，模型只会产生噪音）
# ComfyUI 或StableSwarmUI V-pred 设置 #
添加节点“ModelSamplingDiscrete”，Sampling = “v_prediction” [右键单击\AddNode\advanced\model]
工作流程：[加载CheckPoint]-[ModelSamplingDiscrete]-[KSampler]-[VAE解码]-[保存图像]
# ComfyUI V-pred 设置（旧版）#
1.将配置“.yaml”复制到[ComfyUI\models\configs]并刷新ComfyUI
2.“加载检查点（带配置）” [右键\AddNode\advanced\model_merging]
工作流程：[加载检查点（带配置）]-[KSampler]-[VAE 解码]-[保存图像]
# 一些故障排除 #
如果您使用新版本的 webui-forge 并且检测 v-pred 模型失败。
尝试更新 webui-forge（单击 update.bat）。
当加载 v-pred 模型时，您会在 cmd 控制台中看到此行：
剩余键：dict_keys([‘v_pred’])
# v4.x 有时会出现这种情况，图像会有点烧坏问题#
使用“带有 __wildcards__ 提示的动态提示”和“批次大小 > 1”
只需使用带有参数的相同提示再次生成，它就会恢复正常。
示例设置
Txt2img Setting
- Steps = 30~40
- Sampler = DPM++ 2M Karras, DPM++ SDE Karras, DDIM, UniPC
- CFG scale = 6~8
- Negative embeddings = deformity_v6, bwu ※[SD-WebUI\embeddings]
- Negative embeddings SDXL = ac-neg1, ac-neg2※[SD-WebUI\embeddings]
Hires. fix
- Hires steps = 15
- Denoising strength = 0.45~0.5
- Hires upscaler = 4x-UltraMix_Smooth ※[SD-WebUI\models\ESRGAN]
Img2img Resize
- Scale to = 1.5~2.0
- Denoising strength = 0.45~0.5
- ControlNet0 = softedge_hed, control_v11p_sd15_softedge
- ControlNet0 Weight = 0.4~0.5
- ControlNet0 Pixel Perfect = true
- ControlNet1 = Reference
- ControlNet1 Weight = 0.2~0.3
- ControlNet1 Pixel Perfect = true
SD Setting
- Stop at CLIP layers = 1
- Eta noise seed delta = 0
- SD VAE = kl-f8-anime2 ※[SD-WebUI\models\VAE]
Txt2img Animated Diff ※no “BREAK”, tokens < 75
- Steps = 30~50
- Sampler = DDIM
- CFG = 10~20
- ImageSize = 512~768 ※large image less motion
- model = mm-Stabilized_mid, motionModel_v03-anime
- Number of frames = 10~14
负面提示 v2
unusual anatomy, pixel \(artwork\), mutilated, malformed, watermark, amputee,
restricted palette, mosaic censorship, deformity, sketch, unfinished, monochrome
3D艺术风格
Prompt:
masterpiece, best quality, light, 4k, 2k,
blender \(software\), ray tracing, 3d, unreal engine
写实风格
Prompt:
[:photorealistic, analog style, realistic, photorealism:0.5]
Negative prompt:
[:bwu:0.5]
Recommend Refiner:
Switch at 0.5
| 使用提示： | CLIP SKIP: 2 |
作品参照
by Valerie Hegarty, by Ian McQue, toony (nick wilde) police uniform, jeans, green eyes, three-quarter portrait, big hat, three-quarter view, holding popsicle, sunset, hotel, plant
Negative prompt: unusual anatomy, mutilated, malformed, watermark, amputee, mosaic censorship
Steps: 16, Size: 1024x1024, Seed: 3216871552, Model: YiffyMix v50, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: Euler SGMUniform, CFG scale: 4, Model hash: ea5c599e31
(elden ring), by Derek Gores, by Crookedtrees, anthro male (blaidd \(elden ring\):1.25), werewolf, armor, grey furred cloak, blue eyes, grey body, eye scar, holding big sword, half-length portrait, night, golden grassland, glowing grass, castle, foggy, realistic
Negative prompt: unusual anatomy, mutilated, malformed, watermark, amputee, mosaic censorship
Steps: 16, Size: 1024x1024, Seed: 3190658019, Model: YiffyMix v50, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: Euler SGMUniform, CFG scale: 4, Model hash: ea5c599e31
(by Jotun22, by Winslow Homer), concept art, gate, feral (diablo:1.25), horns, spike, sculpture, demon, glowing, throne room, tourch, foggy, pillar, lava, pentagram, fantasy, cliff, path
Negative prompt: mutilated, malformed, watermark, amputee, mosaic censorship
Steps: 16, Size: 1024x1024, Seed: 2104622239, Model: YiffyMix v50, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: Euler SGMUniform, CFG scale: 4, Model hash: ea5c599e31
使用介绍
v4.4 [2D,3D,真实,分辨率：512~1088,VAE烘焙]
v-pred 版本，不再有黄色/棕色问题。
该版本不需要 CFG Rescale
使用e621标签（无下划线），Artist标签在YiffyMix v2/v3/v4中非常有效
物种/艺术家网格列表（更新 v43+v37）和Furry LoRA/样本/通配符
VAE：WD-KL-F8-Anime2，Furception 校正对比度和降噪 [设置\稳定扩散\SD VAE]
ClipSkip：在CLIP layers=1 停止 (SD=1, NAI=2) [设置\稳定扩散\剪辑跳过]
为新用户推荐“艺术家列表”，在 v3-v4 YiffyMix 版本中使用。
### v4.x是v-pred版本，按照下面的设置###
# A1111 SD-WebUI 或 Forge V-pred 设置 #
1.下载配置文件“.yaml”粘贴到模型旁边。
2.Raname“.yaml”与模型名称相同。 （检查“.yaml”有[参数化：“v”]行）
3.重新启动SD-WebUI（如果配置加载失败，模型只会产生噪音）
# ComfyUI 或StableSwarmUI V-pred 设置 #
添加节点“ModelSamplingDiscrete”，Sampling =“v_prediction”[右键\AddNode\advanced\model]
工作流程：[加载CheckPoint]-[ModelSamplingDiscrete]-[KSampler]-[VAE解码]-[保存图像]
#ComfyUI V-pred 设置（旧版本）#
1.复制配置“.yaml”到[ComfyUI\models\configs]并刷新ComfyUI
2.“加载检查点（带配置）”[右键\AddNode\advanced\model_merging]
工作流程：[加载检查点（带配置）]-[KSampler]-[VAE 解码]-[保存图像]
＃ 一些问题 ＃
如果您使用新版本的 webui-forge 并未能检测到 v-pred model。
尝试更新 webui-forge（单击 update.bat）。
加载 v-pred 模型时，您将在 cmd 控制台中看到这一行：
剩下的键： dict_keys([‘v_pred’])
# 在这种情况下，v4.x 有时会出现一些问题图像#
使用“带有 __wildcards__ 提示的动态提示”和“批量大小 > 1”
只要用参数同样的提示再次生成，就会恢复正常。
设置示例
Txt2img Setting
- Steps = 40~50
- Sampler = DPM++ 2M Karras, DPM++ SDE Karras, DDIM, UniPC
- CFG scale = 5~10
- Negative embeddings = deformity_v6, bwu ※[SD-WebUI\embeddings]
Hires. fix
- Hires steps = 15
- Denoising strength = 0.45~0.5
- Hires upscaler = 4x-UltraMix_Smooth ※[SD-WebUI\models\ESRGAN]
Img2img Resize
- Scale to = 1.5~2.0
- Denoising strength = 0.45~0.5
- ControlNet0 = softedge_hed, control_v11p_sd15_softedge
- ControlNet0 Weight = 0.4~0.5
- ControlNet0 Pixel Perfect = true
- ControlNet1 = Reference
- ControlNet1 Weight = 0.2~0.3
- ControlNet1 Pixel Perfect = true
SD Setting
Txt2img Animated Diff ※no “BREAK”, tokens < 75
- Steps = 30~50
- Sampler = DDIM
- CFG = 10~20
- ImageSize = 512~768 ※large image less motion
- model = mm-Stabilized_mid, motionModel_v03-anime
- Number of frames = 10~14
否定提示 v2
unusual anatomy, pixel \(artwork\), mutilated, malformed, watermark, amputee,
restricted palette, mosaic censorship, deformity, sketch, unfinished, monochrome
3D艺术风格
Prompt:
masterpiece, best quality, light, 4k, 2k, blender \(software\), ray tracing, 3d, unreal engine
写实风格
Prompt:
masterpiece, best quality, light, 4k, 2k, photography,
photorealistic, analog style, realistic, photorealism
Negative prompt:
deformity_v6, bwu, high contrast, over saturated, glossy
| 使用提示： | CLIP SKIP: 1 |
作品参照
(by Kyoto Animation, by Christopher Balaskas, by Glacierclear, by Sususuigi), (duo toony chibi, frisk \(undertale\), asriel dreemurr, hug:1.25), shirt, pants, front view, (three-quarter portrait:1.25), arms above head, holding flower, happy, close-up, BREAK, colorful cave, colorful mountain, foggy, glowing, night, waterfall, crystal, plant, sky, masterpiece, best quality, light, 4k, 2k, photography
Negative prompt: unusual anatomy, mutilated, malformed, watermark, amputee, mosaic censorship, sketch, monochrome
Steps: 40, VAE: Furception.vae.safetensors, Size: 640x896, Seed: 3086791076, Model: YiffyMix v44, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: DPM++ 2M Karras, CFG scale: 7, Model hash: 11621081aa
(by Sskomu \(artist\), by Tekahika, by Jotun22, by Kikurage), (neferpitou:1.3), kemono, yellow eyes, purple shirt, orange knee socks, action pose, hands behind back, (bedroom eyes, looking at viewer, three-quarter portrait, three-quarter view:1.25), BREAK, night, glowing, town, fantasy, plant, cloud, yellow moon, detailed background, foreground, depth of field, ambient silhouette, masterpiece, best quality, light, 4k, 2k, photography
Negative prompt: unusual anatomy, mutilated, malformed, watermark, amputee, mosaic censorship, sketch, monochrome
Steps: 40, VAE: Furception.vae.safetensors, Size: 640x896, Seed: 3324581278, Model: YiffyMix v44, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: DPM++ 2M Karras, CFG scale: 7, Model hash: 11621081aa
(by Albert Marquet, by Kinokoningen, by Mamimi \(artist\), by Yellowroom), antrho female (rito, peafowl:1.3), (blue feathers), beak, winged arms, hands on thighs, looking at viewer, three-quarter view, half-length portrait, (the legend of zelda, blue tribal clothing:1.25), BREAK, colorful jungle, waterfall, japaness town, wooden house, foggy, sky, cloud, detailed background, foreground, depth of field, ambient silhouette, masterpiece, best quality, light, 4k, 2k, photography
Negative prompt: (nipples:1.25), mutilated, malformed, watermark, amputee, mosaic censorship, sketch, monochrome
Steps: 40, VAE: Furception.vae.safetensors, Size: 640x896, Seed: 1101203746, Model: YiffyMix v44, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7, Sampler: DPM++ 2M Karras, CFG scale: 7, Model hash: 11621081aa
