Type | |
Stats | 293 588 |
Reviews | (66) |
Published | Jun 24, 2024 |
Base Model | |
Trigger Words | room-style 3d realistic chibi |
Hash | AutoV2 CA5D59D44E |
有事加群:188546326
Please strictly follow my instructions
请严格按照指示使用
This is a semi-finished commercial commission. Follow my instructions to use it, and it can be directly applied to some extent in your work
这是一个商业委托的半成品,按照我的说明来使用,可一定程度在工作中直接使用
1.This model was trained using a game dataset based on a3.1, but due to resolution issues, the details will be very blurry when the weight is set to 1. Therefore, please keep the weight between 0.5 and 0.7
1.本模型使用游戏数据集基于a3.1训练,但由于分辨率原因,权重为1时,画面细节会非常模糊,所以权重请保持0.5-0.7
2.If you can accurately describe a room, please write the prompt yourself. If you can only describe beautiful girls, then open GPT and follow my instructions.Copy the following text to GPT:
I am using AI to generate a painting of a room. I will give various themes, such as kindergarten, and you need to help me describe the scene. For example, what kind of floor, windows, and furniture should be included, and what items might appear in the room. Do not include any people, just describe the scene. List these items in English, separated by commas.
Then tell GPT the theme, such as a VTuber's room, a Chinese tea house, or a fetish interrogation room, etc.
2.如果你能很好的描述一个房间,请自己写提示词,如果你只会描述美少女,那么打开gpt,按照我说的来,把下面这段话复制给gpt:
我在使用ai绘画描述房间,接下来我会提出各种主题,如,幼儿园,你需要帮我描述场景,如什么样子的地板,窗户,里面要有哪些家具,可能会出哪些物品,但是不要出现人物,只描述一个场景,你帮我列出这些物品的单词,用英文,逗号分隔
然后告诉gpt主题,一个vtuber的房间,一个中式茶馆,或者一个情趣拷问房间等。
3.Suggest using keywords such as 3D, white background, realistic, chibi, room-style. These can be used separately. Generally, adding 3D and realistic makes the image clearer and the furniture structure more reasonable.Since there are no people in the dataset, if you want to add characters, either use traditional prompts with a weight of 1.4 or higher, or directly tell GPT to include character descriptions.
3.提示词中使用3d,white background,realistic,chibi,room-style这些,可以分开使用,通常来说,加入3d和realistic,画面会更清晰,家具结构更合理,由于数据集中没有人,所以如果你想加入人物,或者用传统提示词,但是权重1.4或者更高,或者直接告诉gpt加入人物描写。
4.The resolution is 1365x768. Do not use hires.fix for upscaling. Instead, enter I2I and use Tiled Diffusion for upscaling with a redraw extent of 0.6.
4.分辨率是1365x768,不要使用hires.fix进行放大,进入I2I,使用Tiled Diffusion进行放大,重绘幅度0.6。