一、概述

• 我们更新了Neta Art XL V2.0，在1.0版本的基础上，优化了角色的姿势动态、进一步强化手部稳定性，并且非常擅长用有氛围感的图片讲述故事。
• 主要动机：V1.0中，我们的角色默认生成的姿势比较固定，而新版本对于动态感、多样的角色镜头表达和更加精准的特征控制有所增强。在增加更多风格多样性的基础上，保持了更加稳定的肢体稳定性，尤其是手部现在出现五指（而不是四个指头或者六个）的概率更高了！。进一步强化的Rectified Flow和Noise Offset等问题的训练方法，现在的画面史诗感极强，可以出现很亮和很暗的景色了（如下图）。

•  
•  
•  
•  
• {
•  "prompt": "1girl, full moon, moonlight shadow, cinematic lighting, battle field, fighting armies, castle, warriors around, waving, looking at viewer, Heterochromatic pupil, sitting on a cliff, very dark, epic scene, inferno, cowboy shot, depressed, angel wings, solo, white long hair, wings bangs, beautiful color, amazing quality,",
•  "negative_prompt": "nsfw, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, rating: sensitive, low contrast, signature, flexible deformity, abstract, low contrast, ",
•  "resolution": "1344 x 768",
•  "guidance_scale": 8,
•  "num_inference_steps": 28,
•  "sampler": "Euler a",
•  "use_lora": null,
•  "use_upscaler": {
•    "upscale_method": "bislerp",
•    "upscaler_strength": 0.65,
•    "upscale_by": 1.5,
•    "upscale_cfg": 11,
•    "new_resolution": "2016 x 1152"
•  }
• }

提示指南

在Neta V2中，我们重新基于 ArtiWaifu Diffusion V1.0（AWA1）作为模型进行训练。因此采用ArtiWaifu中总结的提示顺序：

标签顺序:  art style (by xxx) -> character (1 frieren (sousou no frieren)) -> race (elf) -> composition (cowboy shot) -> painting style (impasto) -> theme (fantasy theme) -> main environment (in the forest, at day) -> background (gradient background) -> action (sitting on ground) -> expression (expressionless) -> main characteristics (white hair) -> other characteristics (twintails, green eyes, parted lip) -> clothing (wearing a white dress) -> clothing accessories (frills) -> other items (holding a magic wand) -> secondary environment (grass, sunshine) -> aesthetics (beautiful color, detailed) -> quality (best quality) -> secondary description (birds, cloud, butterfly)

[{

 "prompt": "(oil-pasto:0.6), half-pasto, anime coloring, 1girl, raiden shogun\(genshin_impact\), genshin_impact, solo, smile, sitting beside window, indoor, classroom, hand rest, japanese school uniform,"

},

{

 "prompt": "(line sketch, (black and white, monochrome:1.2), draft, 1girl, solo, march 7th\(honkai: star rail\),honkai: star rail, cowboy shot, medium shot, smile, standing, hand rest,"

},

{

 "prompt": "gufeng style, 2.5d, impasto, old withered vines and ancient trees, a murder of crows descending into darkness, a small bridge over gently flowing water, quaint houses nestled by the river, best quality, masterpiece, highres,"

},

{

 "prompt": "(2.5d:1.4),(pseudo-impasto:1.2),best quality, masterpiece, ultra-detailed, illustration, detailed light,an extremely delicate and beautiful, incredibly_absurdres, <lora:sdxl_zzz_all_v2:1>,1girl, nicole_demara, solo,crop_top, hair_ribbon, two_side_up, black_ribbon, tube_top, midriff, hairclip, bare_shoulders, black_shorts, cleavage, long_sleeves, cowboy shot, standing, medim shot,"

},

{

 "prompt": "pixel art, 1girl, solo, kar98k\(girls' frontline\),girls' frontline, cowboy shot, standing, medim shot,"

}]

负向提示词: (worst quality:1.3), low quality, lowres, messy, abstract, ugly, disfigured, bad anatomy, draft, deformed hands, fused fingers, signature, text, multi views

采样器参数: Eular a 作为默认推荐, 28+ 步及以上的推理。

Neta Art XL 2.0 仍然支持非常广泛的cfg范围（支持3 - 20）。但是我们这次将模型的标准CFG重新对齐到了7-9的常规范围内方便大家替换实验。

六、总结和展望

Neta Art DiT后续训练中。 请继续关注我们，并免费测试我们的产品：http://nieta.art

B站：https://space.bilibili.com/505727005

小红书：https://www.xiaohongshu.com/user/profile/63f2ebf2000000001001e206

推特：https://twitter.com/netaart_ai

Civitai：https://civitai.com/user/neta_art