RunningHub
Search
Invite
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2
Back
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2
CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2

CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2

0 Views0 Likes1 Favorites
realisticconceptkrea 2unknown

NineCrowns

NineCrowns

realisticconceptkrea 2unknown

Model Information

Active
Model Type：
LoRA
Basic Model：
krea2
Trigger Words：
crown_anime2real, A2R, transform the image to realistic photograph
Resource Name：
models/loras/CROWN_Krea2_Edit_Anime2Real_1to1_V1_r64.safetensors
MD5：
d6e1887d95e38b6812238f7e98a80762

🎨➡️📸 CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2 (Turbo & Raw Edit)

[English]
Engineered by NineCrowns ([CROWN]) at native 1024×1536 on an 8×NVIDIA B200 cluster (Rank 64) for Krea 2 (SingleStreamDiT) using dual-pathway reference conditioning (Qwen3-VL-4B-Instruct 12-layer stacked vision-text features + Qwen-Image-VAE index_timestep_zero reference latents with isolate_refs=True).
Built with an orthogonal Hybrid Rank-64 SVD Subspace: ranks 0..31 lock 1:1 pixel-accurate spatial cross-attention (built upon the Apache-2.0 krea2:o_edit_turbo foundation), while ranks 32..63 inject pure real-human East Asian facial bone structure, natural human iris/eyelid proportions, and DSLR/iPhone skin micro-texture:

  1. True Real-Human Anatomy & Skin Pores (crown_anime2real, A2R, transform the image to realistic photograph): Converts flat 2D cel-shaded anime illustrations into authentic live-action photographs with natural human eye proportions, realistic skin pores, subsurface scattering, and individual physical hair strands—eliminating plastic 2.5D cosplay doll faces.
  2. 1:1 Expression, Hairstyle, Outfit & Background Lock: Preserves exact lip state (closed-mouth vs bright smile), bangs/ponytail/bun geometry, collar ribbons, hats, scarves, stethoscopes, and background lighting layout.
  3. 8-Step Krea 2 Turbo Ready: Plug-and-play single .safetensors file in ComfyUI with ComfyUI-Krea2-Ostris-Edit (FluxKontextMultiReferenceLatentMethod = index_timestep_zero + Krea2OstrisEditModelPatch = True).

[中文说明]
由 NineCrowns ([CROWN]) 在 8×NVIDIA B200 集群上以原生 1024×1536 分辨率打造的 Krea 2 (Turbo / Raw) 二次元 1:1 像素级重绘成真人 LoRA（Rank 64）。采用正交双子空间架构（前 32 秩锁死 RoPE axis=1 像素级空间对齐，后 32 秩注入高锐度纯真人东亚骨相与真皮毛孔 SVD 子空间）：

  1. 100% 真人骨相重塑与毛孔肤质：将平涂二次元五官自然重塑为真实人类眼裂比例、鼻梁下颌骨相、微毛孔肤质与根根分明真实发丝，彻底告别 2.5D 假娃娃感。
  2. 1:1 发型、衣领、配饰与表情严丝合缝：精准还原原图刘海发辫、蝴蝶结领口、贝雷帽、围巾、听诊器、嘴唇开合表情与背景光影构图。
  3. 8 步 Krea 2 Turbo 极速直出：单文件标准 ComfyUI diffusion_model.* 格式，配合 Krea 2 Turbo 仅需 8 步（CFG: 1.0, mu: 1.15）即可秒级出图。
📊 Recommended Settings / 推荐推理参数
  • Base Model (底模): krea2_turbo_bf16.safetensors (Recommended, 8 steps) or krea2_raw_bf16.safetensors + Qwen3-VL-4B-Instruct + qwen_image_vae.safetensors
  • Edit Nodes (ComfyUI 节点): FluxKontextMultiReferenceLatentMethod ("index_timestep_zero") + Krea2OstrisEditModelPatch (True)
  • Sampler & Schedule (采样参数): 1024×1536, euler, 8 steps, mu = 1.15 (or beta57), CFG = 1.0
  • LoRA Weight (推荐权重): 0.95 ~ 1.00 (All 8 unseen showcase comparisons generated at 1.00)
  • Trigger Words (触发词): crown_anime2real, A2R, transform the image to realistic photograph

🎨➡️📸 CROWN 2D Anime-to-Real 1:1 Photorealistic Redraw | Krea 2 (Turbo & Raw Edit)

[English]
Engineered by NineCrowns ([CROWN]) at native 1024×1536 on an 8×NVIDIA B200 cluster (Rank 64) for Krea 2 (SingleStreamDiT) using dual-pathway reference conditioning (Qwen3-VL-4B-Instruct 12-layer stacked vision-text features + Qwen-Image-VAE index_timestep_zero reference latents with isolate_refs=True).
Built with an orthogonal Hybrid Rank-64 SVD Subspace: ranks 0..31 lock 1:1 pixel-accurate spatial cross-attention (built upon the Apache-2.0 krea2:o_edit_turbo foundation), while ranks 32..63 inject pure real-human East Asian facial bone structure, natural human iris/eyelid proportions, and DSLR/iPhone skin micro-texture:

  1. True Real-Human Anatomy & Skin Pores (crown_anime2real, A2R, transform the image to realistic photograph): Converts flat 2D cel-shaded anime illustrations into authentic live-action photographs with natural human eye proportions, realistic skin pores, subsurface scattering, and individual physical hair strands—eliminating plastic 2.5D cosplay doll faces.
  2. 1:1 Expression, Hairstyle, Outfit & Background Lock: Preserves exact lip state (closed-mouth vs bright smile), bangs/ponytail/bun geometry, collar ribbons, hats, scarves, stethoscopes, and background lighting layout.
  3. 8-Step Krea 2 Turbo Ready: Plug-and-play single .safetensors file in ComfyUI with ComfyUI-Krea2-Ostris-Edit (FluxKontextMultiReferenceLatentMethod = index_timestep_zero + Krea2OstrisEditModelPatch = True).

[中文说明]
由 NineCrowns ([CROWN]) 在 8×NVIDIA B200 集群上以原生 1024×1536 分辨率打造的 Krea 2 (Turbo / Raw) 二次元 1:1 像素级重绘成真人 LoRA（Rank 64）。采用正交双子空间架构（前 32 秩锁死 RoPE axis=1 像素级空间对齐，后 32 秩注入高锐度纯真人东亚骨相与真皮毛孔 SVD 子空间）：

  1. 100% 真人骨相重塑与毛孔肤质：将平涂二次元五官自然重塑为真实人类眼裂比例、鼻梁下颌骨相、微毛孔肤质与根根分明真实发丝，彻底告别 2.5D 假娃娃感。
  2. 1:1 发型、衣领、配饰与表情严丝合缝：精准还原原图刘海发辫、蝴蝶结领口、贝雷帽、围巾、听诊器、嘴唇开合表情与背景光影构图。
  3. 8 步 Krea 2 Turbo 极速直出：单文件标准 ComfyUI diffusion_model.* 格式，配合 Krea 2 Turbo 仅需 8 步（CFG: 1.0, mu: 1.15）即可秒级出图。
📊 Recommended Settings / 推荐推理参数
  • Base Model (底模): krea2_turbo_bf16.safetensors (Recommended, 8 steps) or krea2_raw_bf16.safetensors + Qwen3-VL-4B-Instruct + qwen_image_vae.safetensors
  • Edit Nodes (ComfyUI 节点): FluxKontextMultiReferenceLatentMethod ("index_timestep_zero") + Krea2OstrisEditModelPatch (True)
  • Sampler & Schedule (采样参数): 1024×1536, euler, 8 steps, mu = 1.15 (or beta57), CFG = 1.0
  • LoRA Weight (推荐权重): 0.95 ~ 1.00 (All 8 unseen showcase comparisons generated at 1.00)
  • Trigger Words (触发词): crown_anime2real, A2R, transform the image to realistic photograph