ComfyUI Extension: Civitai Recipe Finder

Authored by BAIKEMARK

Created

Updated

2 stars

A powerful node suite that finds the community's best 'recipes' for any Civitai model by analyzing top prompts, LoRA triggers, parameters, and combos.

Custom Nodes (0)

    README

    Civitai Recipe Finder

    Description

    Finding the perfect "recipe" — the ideal combination of a model, trigger words, prompts, and generation parameters — is the key to creating stunning AI art. The Civitai Recipe Finder is a powerful suite of ComfyUI nodes designed to uncover these recipes by deeply analyzing data from the Civitai community.

    This node suite moves beyond simple prompt statistics, transforming into a comprehensive analysis toolkit that helps you:

    • Instantly Find Trigger Words: Quickly get official and metadata-based trigger words for any LoRA model.
    • Discover Community Trends: Analyze hundreds of community images to find the most frequently used positive and negative prompts.
    • Uncover Optimal Parameters: Identify the most common generation parameters (Sampler, CFG, Steps, Size, etc.) used by the community for a specific model.
    • Reveal "Golden Combos": Discover which other LoRA models are most frequently used in combination with your selected model, along with their optimal weights.

    The entire suite is built on a modular, pipeline-based philosophy, allowing you to perform quick lookups or build complex, in-depth analysis workflows with maximum efficiency and zero redundant data fetching.

    The Node Suite

    The Recipe Finder is composed of two distinct toolsets to match your specific needs: a lightweight tool for quick lookups, and a powerful pipeline for deep analysis.

    Lightweight Tool (轻量级工具)

    This standalone node is designed for high-frequency, everyday use.

    Lora Trigger Words

    • Purpose: Instantly fetches the two most important sets of trigger words for a LoRA model without any heavy processing.
    • Inputs: lora_name, force_refresh
    • Outputs:

    | Output | Type | Description | | :--- | :--- | :--- | | metadata_triggers | STRING | Trigger words from the local file's metadata (ss_tag_frequency). Reflects the actual training data. | | civitai_triggers | STRING | Official trigger words from the Civitai API (trainedWords). Reflects the author's recommendation. |

    Analyzer Pipeline (分析器流水线)

    This is a powerful, modular pipeline for deep model analysis. The workflow starts with a Fetcher node, which does all the heavy lifting of gathering data once. Its output can then be connected to one or more Analyzer nodes.

    1. Civitai Data Fetcher (CKPT / LORA)

    • Purpose: The engine of the pipeline. It fetches all community image metadata for a given model and outputs it as a single data package. This is the only node in the pipeline that performs heavy network requests.
    • Inputs: model_name, max_pages, sort, retries, timeout, force_refresh
    • Outputs:

    | Output | Type | Description | | :--- | :--- | :--- | | civitai_data | CIVITAI_DATA | A data package containing all fetched raw metadata, ready to be passed to analyzer nodes. | | fetch_summary | STRING | A simple summary of the fetch operation, e.g., "Fetched metadata from 257 images". |

    2. Prompt Analyzer

    • Purpose: Connect to the Fetcher to analyze community prompt usage.
    • Inputs: civitai_data, top_n
    • Outputs:

    | Output | Type | Description | | :--- | :--- | :--- | | positive_prompt | STRING | A ranked list of the most used positive prompts. | | negative_prompt | STRING | A ranked list of the most used negative prompts. |

    3. Parameter Analyzer (CKPT / LORA)

    • Purpose: Connect to the Fetcher to analyze common generation parameters.
    • Inputs: civitai_data
    • Outputs (LORA Version - 9 total):
      • parameter_stats (STRING), top_sampler_name (STRING), top_cfg (FLOAT), top_steps (INT), top_width (INT), top_height (INT), top_hires_upscaler (STRING), top_denoising_strength (FLOAT), top_clip_skip (INT).
    • Outputs (CKPT Version - 10 total):
      • All of the above, plus top_vae_name (STRING).

    4. Resource Analyzer

    • Purpose: Connect to the Fetcher to discover which LoRA models are frequently used together.
    • Inputs: civitai_data
    • Outputs (7 total):
      • associated_resources_stats (STRING) - A formatted summary of the top associated LoRAs and their weights.
      • assoc_lora_1_name (STRING), assoc_lora_1_weight (FLOAT)
      • assoc_lora_2_name (STRING), assoc_lora_2_weight (FLOAT)
      • assoc_lora_3_name (STRING), assoc_lora_3_weight (FLOAT)

    Workflow Examples

    Workflow 1: Quick Trigger Word Lookup

    For when you just need the triggers for a LoRA, instantly.

    [Lora Trigger Words] → [Show Text]

    Workflow 2: Full Model Deep Dive

    Use the pipeline to get a complete picture of a model's community recipe.

                                                         +--> [Prompt Analyzer]
                                                     |
[Civitai Data Fetcher] --> [civitai_data] --+--> [Parameter Analyzer]
                                                     |
                                                     +--> [Resource Analyzer]

    Installation

    1. Place this project under ComfyUI’s custom_nodes directory, for example: 
      ComfyUI/custom_nodes/Civitai_Recipe_Finder
    2. Restart ComfyUI. You will find the new nodes in the menu under the Civitai category and its subcategories.

    Acknowledgement

    The logic for fetching trigger words was inspired by and uses parts of both Extraltodeus/LoadLoraWithTags and idrirap/ComfyUI-Lora-Auto-Trigger-Words. Special thanks to the original authors.

    功能说明

    要创作出惊艳的 AI 艺术作品，关键在于找到完美的“配方 (Recipe)”——即模型、触发词、提示词和生成参数的理想组合。Civitai Recipe Finder 是一套为 ComfyUI 设计的强大节点工具集，旨在通过深度分析 Civitai 社区数据，帮你揭示这些创作配方。

    本节点套件已超越了简单的提示词统计，演变成一个全面的分析工具箱，可以帮你：

    • 即时查找触发词: 快速获取任何 LoRA 模型的官方推荐触发词和元数据触发词。
    • 发现社区趋势: 分析数百张社区图片，找到使用频率最高的正、负向提示词。
    • 揭示最佳参数: 识别社区针对特定模型最常用的生成参数（采样器、CFG、步数、尺寸等）。
    • 发掘“黄金组合”: 发现哪些其他的 LoRA 模型最常与你选择的模型搭配使用，以及它们的最佳权重。

    整个套件基于模块化的“流水线”设计哲学，让你既能进行快速查找，也能构建复杂深度的分析工作流，同时确保了最高效率，杜绝了任何重复的数据获取。

    节点套件说明

    Recipe Finder 由两组独立的工具构成，以匹配你的不同需求：一组用于快速查找的轻量级工具，和一套用于深度分析的强大流水线。

    轻量级工具 (Lightweight Tool)

    这个独立节点专为高频、日常使用场景设计。

    Lora Trigger Words (Lora 触发词)

    • 用途: 无需任何重度处理，即时获取一个 LoRA 模型的两组核心触发词。
    • 输入: lora_name, force_refresh
    • 输出:

    | 输出端口 | 类型 | 说明 | | :--- | :--- | :--- | | metadata_triggers | STRING | 从本地文件元数据 (ss_tag_frequency) 提取的触发词。反映了训练数据。 | | civitai_triggers | STRING | 从 Civitai API (trainedWords) 获取的官方触发词。反映了模型作者的明确推荐。 |

    分析器流水线 (Analyzer Pipeline)

    这是一套为深度模型分析设计的强大、模块化的流水线。工作流由一个 Fetcher (获取器) 节点开始，它负责一次性完成所有重度的数据采集工作，其输出可以连接到一个或多个 Analyzer (分析器) 节点。

    1. Civitai Data Fetcher (CKPT / LORA) (数据获取器)

    • 用途: 流水线的引擎。它为指定的模型获取所有社区图片元数据，并将其打包成一个数据包输出。这是流水线中唯一进行重度网络请求的节点。
    • 输入: model_name, max_pages, sort, retries, timeout, force_refresh
    • 输出:

    | 输出端口 | 类型 | 说明 | | :--- | :--- | :--- | | civitai_data | CIVITAI_DATA | 包含所有已获取的原始元数据的数据包，可供下游的分析器节点使用。 | | fetch_summary | STRING | 对获取操作的简单总结，例如：“已从3个页面获取257张图片的元数据。” |

    2. Prompt Analyzer (提示词分析器)

    • 用途: 连接到 Fetcher 节点，用于分析社区提示词的使用情况。
    • 输入: civitai_data, top_n
    • 输出:

    | 输出端口 | 类型 | 说明 | | :--- | :--- | :--- | | positive_prompt | STRING | 使用频率最高的正向提示词排序列表。 | | negative_prompt | STRING | 使用频率最高的负向提示词排序列表。 |

    3. Parameter Analyzer (CKPT / LORA) (参数分析器)

    • 用途: 连接到 Fetcher 节点，用于分析常见的生成参数。
    • 输入: civitai_data
    • 输出 (LORA 版 - 共9个):
      • parameter_stats (STRING), top_sampler_name (STRING), top_cfg (FLOAT), top_steps (INT), top_width (INT), top_height (INT), top_hires_upscaler (STRING), top_denoising_strength (FLOAT), top_clip_skip (INT)。
    • 输出 (CKPT 版 - 共10个):
      • 包含 LORA 版所有输出，并额外增加 top_vae_name (STRING)。

    4. Resource Analyzer (关联资源分析器)

    • 用途: 连接到 Fetcher 节点，用于发现哪些 LoRA 模型经常被组合使用。
    • 输入: civitai_data
    • 输出 (共7个):
      • associated_resources_stats (STRING) - 格式化好的、关于热门关联 LoRA 及其权重的总结报告。
      • assoc_lora_1_name (STRING), assoc_lora_1_weight (FLOAT)
      • assoc_lora_2_name (STRING), assoc_lora_2_weight (FLOAT)
      • assoc_lora_3_name (STRING), assoc_lora_3_weight (FLOAT)

    工作流示例

    工作流 1: 快速查找触发词

    当你只想立刻知道一个 LoRA 的触发词时使用。

    [Lora Trigger Words] → [Show Text]

    工作流 2: 模型深度挖掘

    使用流水线来全面了解一个模型的社区“配方”。

                                                         +--> [Prompt Analyzer]
                                                     |
[Civitai Data Fetcher] --> [civitai_data] --+--> [Parameter Analyzer]
                                                     |
                                                     +--> [Resource Analyzer]

    安装

    1. 将项目文件夹放入 ComfyUI 的 custom_nodes 目录，例如： 
      ComfyUI/custom_nodes/Civitai_Recipe_Finder
    2. 重启 ComfyUI。你将在 Civitai 菜单及其子菜单中找到所有新节点。

    鸣谢

    本项目中获取触发词的逻辑，其灵感和部分代码实现来源于 Extraltodeus/LoadLoraWithTags 和 idrirap/ComfyUI-Lora-Auto-Trigger-Words。在此特别感谢原作者们。