AI生成图像

碎碎念

耗时三四天左右，前天折腾的AI生成视频现在是完全失败了，原因吗，当然也是很简单的，硬件性能实在是不足了，显存不停的溢出，两张2080ti目前来说的话还是实在是有点不太足够。没辙了，放弃了，等后面看着有更好的显卡再说吧。
现在折腾AI生成图像

使用Fooocus项目
github地址：https://github.com/lllyasviel/Fooocus

环境配置

conda create --name image python=3.10
git clone https://github.com/lllyasviel/Fooocus.git
cd Fooocus
conda env create -f environment.yaml
conda activate fooocus
pip install -r requirements_versions.txt

conda activate fooocus
python entry_with_update.py --listen
#对于 Fooocus 动漫版使用python entry_with_update.py --preset anime或。现实版python entry_with_update.py --preset realistic

当然网络问题是保留节目了
经常会卡在某一个环节如下所示

(image) yanchang@SDFMU3:~/DATA/AIimage/Fooocus$ python launch.py --preset realistic
[System ARGV] ['launch.py', '--preset', 'realistic']
Python 3.10.18 | packaged by conda-forge | (main, Jun  4 2025, 14:45:41) [GCC 13.3.0]
Fooocus version: 2.5.5
Loaded preset: /home/yanchang/DATA/AIimage/Fooocus/presets/realistic.json
[Cleanup] Attempting to delete content of temp dir /tmp/fooocus
[Cleanup] Cleanup successful
Downloading: "https://huggingface.co/lllyasviel/misc/resolve/main/xlvaeapp.pth" to /home/yanchang/DATA/AIimage/Fooocus/models/vae_approx/xlvaeapp.pth

只需要一直用类似于下面的命令行代替每一个需要的模型，不会的可以找ai生成

wget -O /home/yanchang/DATA/AIimage/Fooocus/models/vae_approx/xlvaeapp.pth "https://huggingface.co/lllyasviel/misc/resolve/main/xlvaeapp.pth"

一直到如下，就可以打开网页端进行图像生成了

绕过安全审查

文件一、Fooocus/extras/safety_checker/models/safety_checker.py。复制后替换为

# from https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py

# Copyright 2024 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import numpy as np
import torch
import torch.nn as nn
from transformers import CLIPConfig, CLIPVisionModel, PreTrainedModel
from transformers.utils import logging

logger = logging.get_logger(__name__)


def cosine_distance(image_embeds, text_embeds):
    normalized_image_embeds = nn.functional.normalize(image_embeds)
    normalized_text_embeds = nn.functional.normalize(text_embeds)
    return torch.mm(normalized_image_embeds, normalized_text_embeds.t())


class StableDiffusionSafetyChecker(PreTrainedModel):
    config_class = CLIPConfig
    main_input_name = "clip_input"

    _no_split_modules = ["CLIPEncoderLayer"]

    def __init__(self, config: CLIPConfig):
        super().__init__(config)

        self.vision_model = CLIPVisionModel(config.vision_config)
        self.visual_projection = nn.Linear(config.vision_config.hidden_size, config.projection_dim, bias=False)

        self.concept_embeds = nn.Parameter(torch.ones(17, config.projection_dim), requires_grad=False)
        self.special_care_embeds = nn.Parameter(torch.ones(3, config.projection_dim), requires_grad=False)

        self.concept_embeds_weights = nn.Parameter(torch.ones(17), requires_grad=False)
        self.special_care_embeds_weights = nn.Parameter(torch.ones(3), requires_grad=False)

    @torch.no_grad()
    def forward(self, clip_input, images):
        # 直接返回原始图像，所有图像都被认为是安全的
        if torch.is_tensor(images):
            batch_size = images.shape[0]
        else:
            batch_size = len(images)
        
        has_nsfw_concepts = [False] * batch_size
        return images, has_nsfw_concepts

    @torch.no_grad()
    def forward_onnx(self, clip_input: torch.Tensor, images: torch.Tensor):
        # ONNX版本也同样处理
        batch_size = images.shape[0]
        has_nsfw_concepts = torch.tensor([False] * batch_size, device=images.device)
        return images, has_nsfw_concepts

文件二、Fooocus/extras/censor.py替换为以下内容

import os
import numpy as np
import torch
from transformers import CLIPConfig, CLIPImageProcessor

# 假设以下模块是您项目的一部分，并且可以正确导入
# 如果这些导入引发错误，请确保模块路径正确
import ldm_patched.modules.model_management as model_management
import modules.config
from extras.safety_checker.models.safety_checker import StableDiffusionSafetyChecker
from ldm_patched.modules.model_patcher import ModelPatcher

# 假设 __file__ 变量可用，如果作为独立脚本运行则没有问题
# 如果在交互式环境（如 Jupyter）中运行，您可能需要手动指定路径
try:
    safety_checker_repo_root = os.path.join(os.path.dirname(__file__), 'safety_checker')
except NameError:
    # 如果 __file__ 未定义，提供一个备用路径或当前工作目录
    safety_checker_repo_root = os.path.join('.', 'safety_checker')

config_path = os.path.join(safety_checker_repo_root, "configs", "config.json")
preprocessor_config_path = os.path.join(safety_checker_repo_root, "configs", "preprocessor_config.json")


class Censor:
    """
    一个审查类，用于检查图像是否包含不安全内容。
    此类已被修改以禁用安全检查功能。
    """
    def __init__(self):
        """初始化审查器实例变量。"""
        self.safety_checker_model: ModelPatcher | None = None
        self.clip_image_processor: CLIPImageProcessor | None = None
        self.load_device = torch.device('cpu')
        self.offload_device = torch.device('cpu')

    def init(self):
        """
        （此方法在修改后不再被调用）
        原始功能：加载并初始化安全检查模型。
        """
        if self.safety_checker_model is None and self.clip_image_processor is None:
            safety_checker_model = modules.config.downloading_safety_checker_model()
            self.clip_image_processor = CLIPImageProcessor.from_json_file(preprocessor_config_path)
            clip_config = CLIPConfig.from_json_file(config_path)
            model = StableDiffusionSafetyChecker.from_pretrained(safety_checker_model, config=clip_config)
            model.eval()

            self.load_device = model_management.text_encoder_device()
            self.offload_device = model_management.text_encoder_offload_device()

            model.to(self.offload_device)

            self.safety_checker_model = ModelPatcher(model, load_device=self.load_device, offload_device=self.offload_device)

    # ==================================================================
    # 核心修改部分
    # ==================================================================
    def censor(self, images: list | np.ndarray) -> list | np.ndarray:
        """
        无论输入什么，都直接返回原始图像。
        这有效地禁用了安全检查功能。
        """
        # 不执行任何检查，直接返回收到的图像
        return images
    # ==================================================================


# 创建 Censor 类的实例，并将其 censor 方法赋值给 default_censor
# 这样，其他地方调用 default_censor(...) 时，就会执行我们修改后的逻辑
default_censor = Censor().censor

搭配前几天的换脸，ok