ControlNet Cannyでディテールアップ / DetailUpWithControlNetCanny

NoginowaNoginowa
24
616
controlnetcomfyuiupscaleworkflowcanny工具
v1

详情

下载文件

模型描述

这是一个ComfyUI的示例工作流，先使用t2i，再通过ControlNet Canny进行细节增强，并支持使用UltimateSDUpscale进行上采样。请自由定制使用。

前提条件 / Prerequisites

自定义节点 / Custom nodes

使用了以下自定义节点。

The following custom nodes are used.

使用方法 / How to use

相同内容也记录在工作流中的Note节点中。

The same information is also written in the Note node in the workflow.

  1. 在“Settings & Prompts”组的“SDXL Empty Latent Image (rgthree)”中设置生成图像的尺寸。

  2. 在“Settings & Prompts”组的“Eff. Loader SDXL”中，将“base_ckpt_name”设置为使用的模型，将“base_clip_skip”设置为CLIP Skip值。

  3. 如需使用LoRA，请在“Settings & Prompts”组的“LoRA Stacker”中设置要使用的LoRA。

  4. 在“Settings & Prompts”组的“Eff. Loader SDXL”中输入正向提示词和负向提示词。

  5. 在“Generate”组的“KSampler SDXL (Eff.)”中设置步数、CFG、采样器和调度器。

  6. 在“ControlNet Canny”组的“Load ControlNet Model”中设置要使用的Canny模型。
    如果没有Canny模型，请从 https://huggingface.co/diffusers/controlnet-canny-sdxl-1.0 下载 diffusion_pytorch_model.fp16.safetensors。
    下载后建议重命名文件，以便识别为Canny模型。

  7. 在“ControlNet Canny”组的“Canny Edge”中按喜好调整参数。（如不确定，可保持默认设置）

  8. 在“ControlNet Canny”组的“CLIP Text Encode++ (Positive)”中输入额外的正向提示词。
    可留空，但添加提示词可实现更多调整效果。
    例如，如示例工作流中输入“Light particles”，则会在t2i图像中额外生成光粒子；输入“Satin dress”，则裙子会呈现更光滑的缎面质感。
    输入“Blue texture”会使整张图像偏蓝，输入“Cat ears”则可后期添加猫耳——可能性无限。

  9. 在“ControlNet Canny”组的“CLIP Text Encode++ (Negative)”中输入额外的负向提示词。

  10. 调整“ControlNet Canny”组的“Apply ControlNet (Advanced)”中的strength值。数值越大，ControlNet效果越强。

  11. 在“ControlNet Canny”组的“KSampler (Efficient)”中设置步数、CFG、采样器、调度器和denoise值。denoise值越大，与t2i图像的差异越明显。

  12. 在“Upscale & Save Image”的“Load Upscale Model”中设置要使用的上采样器。

  13. 在“Upscale & Save Image”的“Ultimate SD Upscale”中设置步数、CFG、采样器、调度器、denoise及其他参数。（如不确定，可保持默认设置）

  14. 执行生成。

  1. Set the image size to be generated in “SDXL Empty Latent Image (rgthree)” in the Settings & Prompts group.

  2. In the "Eff. Loader SDXL" in the Settings & Prompts group, set "base_ckpt_name" to the model to use and "base_clip_skip" to the CLIP Skip.

  3. If you use LoRA, set the LoRA you want to use in "LoRA Stacker" in the Settings & Prompts group.

  4. Enter positive and negative prompts in "Eff. Loader SDXL" in the Settings & Prompts group.

  5. Set the number of steps, CFG, sampler, and scheduler in “KSampler SDXL (Eff.)” in the Generate group.

  6. Set the Canny model to be used in “Load ControlNet Model” of the ControlNet Canny group.
    If you don’t have the Canny model, download diffusion_pytorch_model.fp16.safetensors from https://huggingface.co/diffusers/controlnet-canny-sdxl-1.0.
    After downloading, it is convenient to rename the file so that it can be identified as a Canny model.

  7. Adjust the parameters of "Canny Edge" in the ControlNet Canny group to your liking. (If you are not sure, use the settings as they are.)

  8. Enter additional positive prompt to "CLIP Text Encode++ (Positive)" in the ControlNet Canny group.
    You can leave this blank, but you can make various adjustments by entering additional prompts.
    For example, if you enter "Light particles" as in the sample workflow, additional light particles will be generated in the t2i image, and if you enter "Satin dress", the dress will have a more satin texture.
    If you input "Blue texture" the entire image will have a blue tint, if you input "Cat ears" you can add cat ears later; the possibilities are endless.

  9. Enter additional negative prompts to "CLIP Text Encode++ (Negative)" in the ControlNet Canny group.

  10. Adjust the strength of "Apply ControlNet (Advanced)" in the ControlNet Canny group to your liking. The higher the value here, the stronger the effect of ControlNet.

  11. Set the number of steps, CFG, sampler, scheduler, and denoise in the "KSampler (Efficient)" of the ControlNet Canny group. The larger the denoise, the larger the change from t2i.

  12. Set the upscaler to be used for "Load Upscale Model" in Upscale & Save Image.

  13. Set the number of steps, CFG, sampler, scheduler, denoise, and other parameters in "Ultimate SD Upscale" of Upscale & Save Image. (If you are not sure, use the settings as they are.)

  14. Run the generation.

要点 / Point

  • 在上方的“Fast Groups Muter (rgthree)”中将“ControlNet Canny”组和“Upscale & Save Image”组设为“No”，则仅执行t2i，便于重试生成。

  • 因工作流为横向布局，已添加多个“Fast Groups Muter (rgthree)”。点击各组的“->”即可跳转至对应组。

  • If you set the ControlNet Canny group and Upscale & Save Image group to "No" in the "Fast Groups Muter (rgthree)" at the top, only t2i will be executed, making it easier to retry generation.

  • Since the workflow is horizontal, I added multiple "Fast Groups Muter (rgthree)". You can jump to each group by clicking "->" of each group.

英文内容使用Google翻译进行翻译。

此模型生成的图像

未找到图像。