Have a ComfyUI workflow? Get a readable Workflow Doc in your browser.Make a workflow readable.Try it now ↗
A set of custom nodes for ComfyUI
A set of custom nodes for ComfyUI
Package profile
README
Sage Utils is a comprehensive suite of custom nodes and integrated UI features for ComfyUI. It provides:
Sources
3 sourcesSource excerpts
1 excerptSource context: Repo arcum42/ComfyUI_SageUtils
The node suite supports A1111/Civitai metadata formats, while the UI features provide modern, accessible interfaces for AI-assisted workflow creation.
Model information downloaded from Civitai is cached locally in sage_cache_hash.json and sage_cache_info.json for fast access and reporting. These are located in comfyui/user/default/SageUtils/.
Access AI language models directly within ComfyUI for prompt generation, refinement, and creative assistance:
📖 Complete LLM Tab Guide
Wildcard and tag-based system for constructing complex prompts with LLM enhancement:
__category__ syntax for dynamic, randomized prompt generation📖 Complete Prompt Builder Guide
Seamless data flow between all components:
Save Image w/ Added Metadata
An enhanced Save Image node with extra inputs for param_metadata and extra_metadata, allowing you to embed custom metadata under parameters (A1111 style) and extra. Includes switches to control inclusion of standard ComfyUI metadata.
Construct Metadata / Construct Metadata Lite
Nodes for assembling metadata strings from various workflow inputs. The "Lite" version writes a more minimal set of metadata.
Load Checkpoint w/ Metadata
Loads a checkpoint and outputs model_info, including hash and Civitai data. Model info is cached for quick access and reporting.
Load Diffusion Model w/ Metadata
For loading UNET models with metadata support.
**Simple
Nodes in this pack
100 nodesSage_Ace15AudioEncode
Compact description unavailable for this node.
1 input · 8 parameters
Sage_AceAdvOptions
Compact description unavailable for this node.
0 inputs · 6 parameters
Sage_AdvSamplerInfo
Adv Sampler Info Provides advanced sampler settings for granular control over the sampling process. - addnoise (BOOLEAN, required): Whether to add noise during sampling - Default: True - startatstep (INT, required): Step number to start sampling from - Default: 0, Range: 0 to 10,000 - endatstep (INT, required): Step number to end sampling at - Default: 10,000, Range: 0 to 10,000 - returnwithleftovernoise (BOOLEAN, required): Whether to return result with remaining noise - Default: False - ADVSAMPLERINFO: Dictionary containing advanced sampler parameters Use when you need granular control over the sampling process beyond basic parameters. Connect to KSampler nodes to enable advanced sampling features like partial denoising, step ranges, and noise control. - Designed to work with SageKSampler and related sampler nodes - addnoise: Set to False for img2img workflows where noise should not be added - startatstep and endatstep: Enable partial sampling for multi-stage workflows - returnwithleftovernoise: Useful for multi-pass sampling or when preserving noise structure - Step range allows for sampling only specific portions of the denoising process - Provides functionality similar to ComfyUI's advanced KSampler - Optional input for sampler nodes - basic functionality works without it
0 inputs · 4 parameters
Sage_AnythingToStr
Anything to String Converts any incoming value to a string and outputs it without rendering preview text in the UI. - Sage Utils/text/util - any (ANY): Any input value, including strings, numbers, lists, or custom types - STRING: String representation of the input value Use this node when you need string conversion for downstream nodes but do not want an output display node. This node: - Accepts any input type - Converts non-list values with standard string conversion - Converts lists by joining each item as a line - Outputs only the string value (no preview panel) - Unlike SageViewAnything, this node is not an output display node - For list inputs, items are joined with newline separators
Sage_AverageConditioning
Compact description unavailable for this node.
Sage_CheckLorasForUpdates
Check Loras for Updates Checks LoRAs in your stack for available updates from Civitai and provides download information. - lorastack (LORASTACK): The stack of LoRAs to check for updates - force (BOOLEAN): Force a check for updates, even if the LoRA is up to date (default: False) - lorastack (LORASTACK): The original LoRA stack (passed through) - path (STRING): List of local paths for LoRAs that have updates available - latesturl (STRING): List of Civitai URLs for the latest versions of updated LoRAs Use to keep your LoRA library up to date with the latest versions. The node doesn't automatically download updates but provides the information needed to manually update. - Checks each LoRA in the stack against Civitai for newer versions - Only reports LoRAs that have updates available - Force option will re-check even recently checked LoRAs - Returns Civitai URLs in format: https://civitai.com/models/{modelId}?modelVersionId={versionId} - Does not automatically download or replace LoRAs - Useful for maintenance workflows and keeping models current - Requires LoRAs to have Civitai metadata for update checking
0 inputs · 1 parameter
Sage_CheckpointSelector
Checkpoint Selector Selects a model checkpoint and returns its metadata information without loading the actual model. - ckptname (STRING): The name of the checkpoint (model) to select from available checkpoints - modelinfo (MODELINFO): The model path and hash, all in one output Use to choose which model checkpoint to reference in your workflow without loading it. Useful for metadata workflows or conditional model loading. - Does not load the actual model, only returns metadata - Automatically calculates and caches the model hash - Pulls Civitai metadata information for the selected model - Updates the timestamp for the selected model in the cache - Useful for workflows that need model information before deciding whether to load
0 inputs · 1 parameter
Sage_ChromaCLIPLoaderFromInfo
Load Chroma CLIP model component from model info (adds T5 tokenizer options).
Sage_CleanText
Clean Text Cleans or normalizes a text string by removing unwanted characters, excess whitespace, and formatting inconsistencies. Useful for preparing text for prompt or metadata processing. - str (STRING): The string to clean and normalize - cleanedstring (STRING): The cleaned and normalized text Use before passing text to prompt or metadata nodes to ensure clean input. This node applies text cleaning functions that remove extra spaces, normalize line breaks, and clean up formatting issues that can interfere with text processing.
0 inputs · 1 parameter
Sage_CLIPLoaderFromInfo
Load CLIP model component from model info.
Sage_CLIPSelector
Compact description unavailable for this node.
0 inputs · 2 parameters
Sage_CollectKeywordsFromLoraStack
Lora Stack → Keywords Extracts Civitai keywords from a LoRA stack and combines them into a single string. - lorastack (LORASTACK): The stack of LoRAs to extract keywords from - keywords (STRING): Combined keywords from all LoRAs in the stack Use to extract and use keywords for metadata or prompt construction. Place at the end of a LoRA stack, or you won't get keywords for the entire stack. - Goes through each model in the LoRA stack and grabs keywords from Civitai - Combines all keywords into one string for easy use - Returns empty string if no LoRA stack is provided - Must be connected after all LoRAs have been added to the stack - Useful for automatic prompt enhancement based on LoRA metadata - Keywords are sourced from Civitai model information
Sage_CombineCLIPMultilineTextEncode
Compact description unavailable for this node.
0 inputs · 1 parameter
Sage_CombineCLIPTextEncode
Compact description unavailable for this node.
Sage_CombineConditioning
Compact description unavailable for this node.
Sage_ConstructLLMPrompt
Construct LLM Prompt Builds a comprehensive prompt for use with LLM nodes based on predefined templates and customizable options. - prompt (STRING): Base prompt template from available categories (multiline) - extrainstructions (STRING): Additional custom instructions to append (multiline, default: "") - Various style/quality/content options (BOOLEAN): Dynamic options based on available prompt extras (style, quality, contentfocus categories) - prompt (STRING): The constructed prompt ready for LLM use Use to assemble comprehensive prompts for LLM nodes in your workflow. Select from predefined prompt templates and add custom instructions and style options. - EXPERIMENTAL: This node is experimental and may change in future versions - Prompt templates are loaded from llmprompts.json configuration - Available prompts are organized by category and shown as "category/promptname" - Style, quality, and content focus options are dynamically generated based on configuration - Ensures proper sentence-ending punctuation and formatting - Extra instructions are appended after the base prompt and options - Boolean options add their associated prompt text when enabled - Prompt cannot be empty after construction
Sage_ConstructLLMPromptExtra
Construct LLM Prompt Extra Builds extra instructions for LLM prompts using advanced options not covered in the main prompt constructor. - extrainstructions (STRING): Custom extra instructions to include (multiline, default: "") - Various advanced options (BOOLEAN): Dynamic options for categories other than style, quality, and contentfocus - extra (STRING): The constructed extra instructions string Use to assemble advanced extra instructions for LLM nodes. This node provides access to additional prompt options not available in the main constructor. - EXPERIMENTAL: This node is experimental and may change in future versions - Complements SageConstructLLMPrompt by handling advanced options - Options are loaded from llmprompts.json configuration - Only shows options NOT in style, quality, or contentfocus categories - Each enabled boolean option adds its associated prompt text - Custom extrainstructions are prepended to the generated options - Results can be combined with main prompt outputs - Automatic formatting with proper line breaks and spacing
Sage_ConstructMetadataFlexible
Flexible metadata constructor with multiple style options.
0 inputs · 5 parameters
Sage_CropImage
Compact description unavailable for this node.
1 input · 4 parameters
Sage_CubiqImageResize
An advanced image resizing node based on ComfyUI-Essentials functionality. Provides high-quality image resizing with multiple interpolation methods and precise dimension control. - image (IMAGE): The input image to resize - width (INT): Target width in pixels (default: 1024, range: 0 to MAXRESOLUTION) - height (INT): Target height in pixels (default: 1024, range: 0 to MAXRESOLUTION) - interpolation (COMBO): Interpolation method - "nearest", "bilinear", "bicubic", "area", "nearest-exact", "lanczos", "bislerp" - method (COMBO): Resize method - "stretch", "keep proportion", "fill / crop", "pad" - condition (COMBO): When to resize - "always", "downscale if bigger", "upscale if smaller", "if bigger area", "if smaller area" - multipleof (INT): Ensure dimensions are multiples of this value (default: 0, range: 0-1024) - IMAGE (IMAGE): The resized image - width (INT): Actual output width - height (INT): Actual output height - stretch: Stretch image to exact dimensions (may distort aspect ratio) - keep proportion: Maintain aspect ratio, fit within target dimensions - fill / crop: Fill target dimensions, crop excess while maintaining aspect ratio - pad: Fit within dimensions and pad with black borders if needed - nearest: Fastest, pixelated results for pixel art - bilinear: Good balance of speed and quality - bicubic: Higher quality, smoother results - area: Best for downscaling operations - nearest-exact: Precise nearest neighbor - lanczos: Highest quality, best for upscaling - bislerp: ComfyUI's optimized bicubic interpolation - always: Always resize regardless of current size - do…
1 input · 6 parameters
Sage_DualCLIPSelector
Compact description unavailable for this node.
0 inputs · 3 parameters
Sage_DualCLIPTextEncode
Dual Prompt Encode Encodes positive and negative prompts using a CLIP model, returning both conditionings and the input text with optional text cleaning. - clip (CLIP): The CLIP model used for encoding the text - clean (BOOLEAN): Clean up the text, getting rid of extra spaces, commas, etc. (default: False) - pos (STRING): The positive prompt's text (multiline, supports dynamic prompts) - neg (STRING): The negative prompt's text (multiline, supports dynamic prompts) - positive (CONDITIONING): Conditioning containing the embedded positive text used to guide the diffusion model - negative (CONDITIONING): Conditioning containing the embedded negative text. If neg is not connected, it'll be automatically zeroed - postext (STRING): The positive prompt text (cleaned if clean=True) - negtext (STRING): The negative prompt text (cleaned if clean=True) Use to generate conditioning for both positive and negative prompts in workflows. Saves space over two separate CLIP Text Encoders, and automatically zeros any input not connected. - More efficient than using two separate CLIP Text Encoder nodes - Automatically handles unconnected inputs by zeroing the conditioning - Text cleaning removes extra spaces, commas, and formatting issues - Supports dynamic prompts for both positive and negative inputs - Progress bar shows encoding progress for both prompts - Part of "Sage Utils/clip" category for conditioning operations - Returns both the conditioning tensors and the original text for further processing
Sage_DualCLIPTextEncodeLumina2
Dual Prompt Encode (Lumina 2) Encodes positive and negative prompts using a CLIP model with Lumina 2-specific system prompts for enhanced image-text alignment. - clip (CLIP): The CLIP model used for encoding the text - systemprompt (STRING): Lumina 2 system prompt type selection - clean (BOOLEAN): Clean up the text, getting rid of extra spaces, commas, etc. (default: False) - pos (STRING): The positive prompt's text (multiline, supports dynamic prompts) - neg (STRING): The negative prompt's text (multiline, supports dynamic prompts) - poscond (CONDITIONING): Conditioning containing the embedded positive text with system prompt - negcond (CONDITIONING): Conditioning containing the embedded negative text with system prompt. If neg is not connected, it'll be automatically zeroed - postext (STRING): The positive prompt text with system prompt prefix (cleaned if clean=True) - negtext (STRING): The negative prompt text with system prompt prefix (cleaned if clean=True) "You are an assistant designed to generate superior images with the superior degree of image-text alignment based on textual prompts or user prompts." "You are an assistant designed to generate high-quality images with the highest degree of image-text alignment based on textual prompts." Use for workflows that require Lumina 2 prompt encoding with specialized system prompts. The system prompts are automatically prepended to your prompts with delimiter. - Specifically designed for Lumina 2 model compatibility - System prompts enhance image-text alignment quality - Automatically formats prompts as: {systemprompt} {us…
Sage_DualCLIPTextEncodeQwen
Compact description unavailable for this node.
Sage_DynamicJoinText
Compact description unavailable for this node.
Sage_EmptyAceStep15LatentAudio
Compact description unavailable for this node.
0 inputs · 2 parameters
Sage_EmptyLatentImagePassthrough
Empty Latent Passthrough Creates an empty latent image and passes width/height for easier wiring. Includes a switch for SD3 compatibility. - width (INT): The width of the latent images in pixels (default: 1024, range: 16 to MAXRESOLUTION, step: 8) - height (INT): The height of the latent images in pixels (default: 1024, range: 16 to MAXRESOLUTION, step: 8) - batchsize (INT): The number of latent images in the batch (default: 1, range: 1-4096) - sd3 (BOOLEAN): Enable for SD3 compatibility - changes latent channels from 4 to 16 (default: False) - latent (LATENT): The empty latent image batch - width (INT): Pass through the image width - height (INT): Pass through the image height Use to generate latent images and pass dimensions to downstream nodes. The passthrough outputs make it easier to wire width/height to other nodes without additional connections. - Creates empty latent tensors on intermediate device for memory efficiency - SD3 mode uses 16 channels instead of 4 for compatibility with SD3 models - Latent dimensions are automatically scaled down by factor of 8 (standard for diffusion models) - Width and height must be multiples of 8 for proper latent space alignment - Batch size allows creating multiple latent images in a single tensor - Useful for starting generation workflows with proper dimension tracking
0 inputs · 4 parameters
Sage_EnhancerPrompt
Compact description unavailable for this node.
Sage_ErniePromptEnhancerPrompt
Compact description unavailable for this node.
0 inputs · 3 parameters
Sage_FlexibleCLIPSelector
Compact description unavailable for this node.
0 inputs · 1 parameter
Sage_FreeMemory
Free up memory by unloading all models and clearing caches.
Sage_FreeU2
Get the free_u2 settings to apply to the model.
0 inputs · 5 parameters
Sage_GuessResolutionByRatio
Analyzes input dimensions and suggests the closest standard resolution from common aspect ratios, rounded to multiples of 64. - width (INT): Input width dimension (range: 64-8192) - height (INT): Input height dimension (range: 64-8192) - width (INT): Suggested width dimension - height (INT): Suggested height dimension Based on the input width and height, this node guesses a resolution that matches one of the common aspect ratios. The output is rounded to the nearest multiple of 64. - Supports 11 common aspect ratios including 1:1, 5:12, 9:16, 10:16, 5:7, 2:3, 3:4, 4:7, 7:9, 8:10, and 13:19 - Automatically detects landscape vs portrait orientation - Finds the closest aspect ratio by comparing input ratio to predefined ratios - All outputs are rounded to nearest multiple of 64 for optimal AI generation - Swaps dimensions appropriately to maintain landscape/portrait orientation - Defaults to 1024x1024 if no close match is found - Useful for standardizing arbitrary input dimensions to known good ratios - Optimized for AI image generation workflows requiring specific aspect ratios
0 inputs · 2 parameters
Sage_HiDreamE1_Instruction
HiDreamE1 Instruction Generates a formatted prompt for HiDream E1 models based on instruction and description inputs. - instruction (STRING, required): The instruction text for the model - description (STRING, required): The description text for the model - prompt (STRING): Formatted prompt combining instruction and description Use to create properly formatted prompts for HiDream E1 models. The node takes separate instruction and description inputs and combines them into the specific format expected by HiDream E1. - Both instruction and description inputs are required and cannot be empty - Text inputs are automatically cleaned (whitespace trimmed, etc.) - If the instruction doesn't end with a period, one is automatically added - Output format: "Instruction: {instruction}\nDescription: {description}" - Designed specifically for HiDream E1 model prompt formatting requirements
0 inputs · 2 parameters
Sage_JoinText
Join Text Joins two text strings together with a customizable separator. Useful for building prompts or metadata with controlled formatting. - separator (STRING): The separator to use between strings (default: ', ') - addseparatortoend (BOOLEAN): Whether to add separator to the end of the joined string (default: False) - str1 (STRING): First string to join - str2 (STRING): Second string to join - str (STRING): The joined string with separator Use to concatenate text for prompts, metadata, or other string fields. The separator parameter allows you to control how the strings are joined (comma, space, newline, etc.), and the addseparatortoend option lets you append the separator to the final result.
0 inputs · 4 parameters
Sage_KSampler
KSampler w/ Sampler Info A KSampler node designed to work with Sampler Info nodes for streamlined configuration and workflow integration. - model (MODEL): The model used for denoising the input latent - samplerinfo (SAMPLERINFO): Sampler settings from SageSamplerInfo node - positive (CONDITIONING): Conditioning describing attributes to include in the image - negative (CONDITIONING): Conditioning describing attributes to exclude from the image - latentimage (LATENT): The latent image to denoise - denoise (FLOAT): Amount of denoising applied - Default: 1.0, Range: 0.0 to 1.0, Step: 0.01 - advancedinfo (ADVSAMPLERINFO): Advanced sampler options from SageAdvSamplerInfo node - LATENT: The denoised latent image Use for workflows that require flexible sampler configuration and metadata integration. Connect a SageSamplerInfo node to provide centralized sampler settings, and optionally connect SageAdvSamplerInfo for advanced control options. - Designed to work with SageSamplerInfo for centralized configuration - Supports advanced sampling options when SageAdvSamplerInfo is connected - Lower denoise values maintain structure for image-to-image workflows - Sampler info should be routed to both this node and metadata construction - Uses ComfyUI's commonksampler internally for compatibility - Advanced info enables step ranges, noise control, and leftover noise options - Streamlines workflow by separating sampler configuration from execution
4 inputs · 1 parameter
Sage_KSamplerAudioDecoder
KSampler + Audio Decoder A specialized KSampler node designed for audio generation workflows. It performs sampling on latent audio data and automatically decodes it to audio format, outputting both the denoised latent and decoded audio. - model (MODEL): The model used for denoising the input latent - samplerinfo (SAMPLERINFO): Most of the KSampler options. Should be piped both here and to the Construct Metadata node - positive (CONDITIONING): The conditioning describing the attributes you want to include in the audio - negative (CONDITIONING): The conditioning describing the attributes you want to exclude from the audio - latentaudio (LATENT): The latent audio to denoise - vae (VAE): The VAE used for decoding the latent audio - denoise (FLOAT): The amount of denoising applied, lower values will maintain the structure of the initial audio allowing for audio to audio sampling (default: 1.0, range: 0.0-1.0) - advancedinfo (ADVSAMPLERINFO): Optional. Adds in the options an advanced KSampler would have - LATENT: The denoised latent - AUDIO: The decoded audio (44.1kHz sample rate with normalized waveform) This node is specifically designed for audio generation workflows using latent diffusion models. It combines the sampling and decoding steps into a single node for convenience. Key features: - Automatic audio normalization (scales by 5x standard deviation, minimum 1.0) - Fixed 44.1kHz sample rate output - Supports advanced sampling options when connected to SageAdvSamplerInfo - Works with the SageSamplerInfo node for consistent workflow integration - The audio output includes a…
5 inputs · 1 parameter
Sage_KSamplerTiledDecoder
KSampler + Tiled Decoder A KSampler node with integrated VAE decoder that supports optional tiled decoding for memory-efficient processing of large images. - model (MODEL): The model used for denoising the input latent - samplerinfo (SAMPLERINFO): Sampler settings from SageSamplerInfo node - positive (CONDITIONING): Conditioning describing attributes to include in the image - negative (CONDITIONING): Conditioning describing attributes to exclude from the image - latentimage (LATENT): The latent image to denoise - vae (VAE): The VAE used for decoding the latent image - denoise (FLOAT): Amount of denoising applied - Default: 1.0, Range: 0.0 to 1.0, Step: 0.01 - tilinginfo (TILINGINFO): Tiling parameters from SageTilingInfo node for memory-efficient decoding - advancedinfo (ADVSAMPLERINFO): Advanced sampler options from SageAdvSamplerInfo node - LATENT: The denoised latent image - IMAGE: The decoded image (tiled if tilinginfo provided) Use for workflows that require both sampling and immediate image output with optional memory-efficient tiled decoding. Connect SageTilingInfo to enable tiled VAE decoding for large images, and optionally connect SageAdvSamplerInfo for advanced sampling control. - Combines sampling and VAE decoding in a single node - Automatically uses tiled decoding when SageTilingInfo is connected - Tiling enables processing of high-resolution images with limited VRAM - Supports both spatial and temporal tiling for video content - Advanced sampling options available through SageAdvSamplerInfo - Automatically adjusts tiling parameters for compatibility - Handle…
5 inputs · 1 parameter
Sage_LLMPromptText
Unified text generation node that switches between Ollama, LM Studio, and Native providers.
0 inputs · 3 parameters
Sage_LLMPromptVision
Unified vision generation node that switches between REST/OpenAI and Native providers.
1 input · 3 parameters
Sage_LLMPromptVisionRefine
Unified vision-refine node that performs initial generation plus a refinement pass per selected provider.
1 input · 5 parameters
Sage_LMStudioLLMPromptText
Sends a text prompt to an LM Studio LLM and returns the response. - prompt (STRING): The text prompt to send to the model (multiline, default: detailed description prompt) - model (STRING): The LM Studio model to use (from available installed models) - seed (INT): Seed for random number generation (default: 0, range: 0 to 2^32-1) - loadforseconds (INT): Time in seconds to keep model loaded, -1 for indefinitely (default: 0, range: -1 to 3600) - response (STRING): The LLM's text response to the prompt Use to generate text completions or responses from an LM Studio LLM in your workflow. Requires LM Studio to be running with models available. - EXPERIMENTAL: This node is experimental and may change in future versions - Requires LM Studio to be installed and running - Model list is populated from available LM Studio models - Seed ensures reproducible results when set to same value - loadforseconds controls how long model stays in memory after generation - Default prompt is optimized for detailed descriptions - Currently has Ollama dependency in error handling (may be implementation bug) - Supports all text-based LM Studio models
0 inputs · 4 parameters
Sage_LMStudioLLMPromptVision
Sends a vision prompt with image input to an LM Studio LLM and returns the response. - prompt (STRING): The text prompt to send to the model (multiline, default: detailed image description prompt) - model (STRING): The LM Studio vision model to use (from available installed vision models) - image (IMAGE): The image to analyze and describe - seed (INT): Seed for random number generation (default: 0, range: 0 to 2^32-1) - loadforseconds (INT): Time in seconds to keep model loaded, -1 for indefinitely (default: 0, range: -1 to 3600) - response (STRING): The LLM's text response describing the image Use to generate vision-based completions or responses from an LM Studio LLM in your workflow. Ideal for image captioning, analysis, and description tasks. - EXPERIMENTAL: This node is experimental and may change in future versions - Requires LM Studio to be installed and running - Only shows vision-capable models in the dropdown - Image input is required for vision models - Seed ensures reproducible results when set to same value - loadforseconds controls how long model stays in memory after generation - Default prompt is optimized for detailed image descriptions suitable for AI generators - Currently has Ollama dependency in error handling (may be implementation bug) - Supports multimodal input (image + text prompt)
1 input · 4 parameters
Sage_LMStudioLLMPromptVisionRefine
An advanced LM Studio LLM node that provides refined vision-based prompt generation with two-stage processing. This node first generates an initial description using a vision model, then refines that description using a separate text model for enhanced quality and detail. - prompt (STRING): Initial prompt for the vision model (default: detailed image description prompt) - model (COMBO): Available LM Studio vision model for initial analysis - image (IMAGE): The input image to analyze - seed (INT): Random seed for the initial vision model (0 to 2^32-1) - refineprompt (STRING): Prompt for refining the initial description (default: rewrite to be more vivid and detailed) - refinemodel (COMBO): Available LM Studio text model for refinement - refineseed (INT): Random seed for the refinement model (0 to 2^32-1) - initialresponse (STRING): The original response from the vision model - refinedresponse (STRING): The enhanced response after refinement - Stage 1: Vision model analyzes the image using the initial prompt - Stage 2: Text model refines the vision output using the refine prompt - Dual seeding: Separate seeds for reproducible results in both stages - Model flexibility: Can use different models for vision and refinement tasks This node is perfect for: - Creating high-quality prompt descriptions from images with iterative improvement - Generating detailed image analysis that's then enhanced for clarity and engagement - Building sophisticated prompts through a two-model approach - Achieving consistent, reproducible results with separate seeding control 1. Vision model analyzes…
1 input · 6 parameters
Sage_Load_Dataset_From_Folder
Loads a dataset of images and captions from a directory for training.
0 inputs · 4 parameters
Sage_LoadImage
Load Image w/ Size & Metadata Loads an image and outputs its size and embedded metadata along with the image and mask. - image (STRING): The image file to load from the input directory (supports image upload) - image (IMAGE): The loaded image - mask (MASK): The alpha channel mask (if present) - width (INT): The image width in pixels - height (INT): The image height in pixels - metadata (STRING): Embedded metadata from the image file Use to inspect images and extract their metadata for use in workflows. Provides both the image data and useful information about dimensions and embedded metadata. - Automatically scans all files in the ComfyUI input directory - Supports image upload functionality - Extracts alpha channel as mask if present - Metadata includes EXIF data, PNG text chunks, and other embedded information - File validation ensures only valid image files are processed - Uses SHA256 hash to detect file changes for cache invalidation - Supports all common image formats (PNG, JPEG, WebP, etc.) - Useful for loading reference images and extracting generation parameters
Sage_LoadModelFromInfo
Load model components from model info.
Sage_LogicalSwitch
Switch Selects between two inputs based on a boolean condition. Useful for conditional logic in workflows. - condition (BOOLEAN, required): Boolean value to determine which input to select - truevalue (ANY, required): Value to return when condition is true - falsevalue (ANY, required): Value to return when condition is false - result (ANY): The selected value based on the condition Use to route data or control flow based on conditions. When the condition is true, the node outputs truevalue; when false, it outputs falsevalue. This enables conditional branching in workflows. - Accepts any data type for both value inputs (ANY type) - Simple if-else logic: condition ? truevalue : falsevalue - Useful for creating conditional workflows and dynamic data routing - Both value inputs are required even if only one may be used
Sage_LoraStack
Simple Lora Stack Builds and manages a stack of LoRAs with toggles and weights. - enabled (BOOLEAN): Whether to add this LoRA to the stack (default: True) - loraname (STRING): The name of the LoRA from available LoRAs - modelweight (FLOAT): How strongly to modify the diffusion model. This value can be negative (default: 1.0, range: -100.0 to 100.0) - clipweight (FLOAT): How strongly to modify the CLIP model. This value can be negative (default: 1.0, range: -100.0 to 100.0) - lorastack (LORASTACK): Existing LoRA stack to add this LoRA to - lorastack (LORASTACK): The updated LoRA stack Chain multiple nodes for complex LoRA stacking. Connect to LoRA Stack Loader or metadata nodes. If enabled is False, the LoRA is not added to the stack. - Compatible with other node packs that have lorastacks - Negative weights can be used to subtract the LoRA effect - Each LoRA can be individually enabled/disabled - Can be chained together to build complex LoRA combinations - Shows all available LoRAs in the dropdown
0 inputs · 4 parameters
Sage_LoraStackInfoDisplay
Display information for all LoRAs in a lora_stack.
Sage_LoraStackLoader
Lora Stack Loader Loads all LoRAs in a stack and applies them to the model and CLIP, with optional model shifts. - model (MODEL): The diffusion model the LoRA will be applied to - clip (CLIP): The CLIP model the LoRA will be applied to - lorastack (LORASTACK): The stack of LoRAs to load and apply - modelshifts (MODELSHIFTS): The model shifts & FreeU2 settings to apply to the model - model (MODEL): The modified diffusion model with LoRAs applied - clip (CLIP): The modified CLIP model with LoRAs applied - lorastack (LORASTACK): The stack of LoRAs (passed through) - keywords (STRING): Keywords from the LoRA stack Connect to a LoRA stack node to load all LoRAs at once. Supports model shifts for advanced model modifications. - Applies all LoRAs in the stack to both model and CLIP - Shows progress bar during loading process - Extracts and returns keywords from all LoRAs in the stack - Supports model shifts including discrete flow sampling and FreeU v2 - Model shifts can apply x1 or x1000 multipliers for different model types - FreeU v2 can be enabled for improved quality with specific parameters - Compatible with LoRA stacks from other node packs
2 inputs · 0 parameters
Sage_ModelInfo
Model Info Retrieves Civitai model information and metadata based on model info input. - modelinfo (MODELINFO, required): Model information containing hash and metadata - basemodel (STRING): The base model type (e.g., SDXL, SD1.5, etc.) - name (STRING): Model name with version information - url (STRING): Civitai URL for the specific model version - latesturl (STRING): Civitai URL for the latest version of the model - image (IMAGE): Preview image from Civitai Use to retrieve and display comprehensive model information from Civitai. The node queries Civitai's API using the model hash to fetch metadata, URLs, and preview images for documentation or workflow information display. - Requires internet connection to fetch data from Civitai - Uses model hash to query Civitai API for model information - Returns empty strings and blank image if model info is unavailable - Handles exceptions gracefully by returning empty values - The "latesturl" provides a link to the most recent version of the model - Preview image is fetched from Civitai's image URLs - Useful for model documentation and workflow metadata
Sage_ModelInfoDisplay
Display model info as markdown with Civitai details.
Sage_ModelLoraStackLoader
Model + Lora Stack Loader Loads both a checkpoint and a LoRA stack in one node, with optional model shifts. - modelinfo (MODELINFO): The diffusion model info to load. Should be from the checkpoint info node, not a loader node, to avoid loading the model twice - lorastack (LORASTACK): The stack of LoRAs to load and apply - modelshifts (MODELSHIFTS): The model shifts & FreeU2 settings to apply to the model - model (MODEL): The modified diffusion model with LoRAs applied - clip (CLIP): The modified CLIP model with LoRAs applied - vae (VAE): The VAE model from the checkpoint - lorastack (LORASTACK): The stack of LoRAs (passed through) - keywords (STRING): Keywords from the LoRA stack Use to streamline workflows that require both a model and a LoRA stack. Loads the checkpoint and applies all LoRAs in a single efficient operation. - Inherits functionality from SageLoraStackLoader - Loads model, CLIP, and VAE from checkpoint using modelinfo - Applies all LoRAs in the stack to the loaded model and CLIP - Shows progress bar during loading process - Extracts and returns keywords from all LoRAs in the stack - Supports model shifts and FreeU v2 settings - More efficient than separate model loading and LoRA application - Requires checkpoint modelinfo (not UNET-only models) - Part of "Sage Utils/model" category for complete model loading
Sage_ModelShiftOnly
Get the model shifts to apply to the model.
0 inputs · 2 parameters
Sage_ModelShifts
Model Shifts Configures model shift parameters and FreeU v2 settings for advanced model modifications. - shifttype (STRING): The type of shift to apply to the model - "None", "x1", or "x1000" (x1 for most models, x1000 for Auraflow and Lumina2) - shift (FLOAT): The shift value to apply (default: 3.0, range: 0.0-100.0) - freeuv2 (BOOLEAN): Whether to enable FreeU v2 (default: False) - b1 (FLOAT): FreeU v2 backbone factor 1 (default: 1.3, range: 0.0-10.0) - b2 (FLOAT): FreeU v2 backbone factor 2 (default: 1.4, range: 0.0-10.0) - s1 (FLOAT): FreeU v2 skip factor 1 (default: 0.9, range: 0.0-10.0) - s2 (FLOAT): FreeU v2 skip factor 2 (default: 0.2, range: 0.0-10.0) - modelshifts (MODELSHIFTS): Configuration object containing all shift and FreeU settings Use in workflows that require dynamic model parameter adjustment. Connect to LoRA Stack Loader or Model LoRA Stack Loader nodes. - Model shifts adjust sampling parameters for different model architectures - x1 shift type uses 1000.0 multiplier (for most models) - x1000 shift type uses 1.0 multiplier (for Auraflow and Lumina2) - FreeU v2 can improve generation quality with proper parameter tuning - Settings are passed to model loader nodes for application - "None" shift type disables model shifting - Used by the model loader node for advanced model modifications - Part of the "Sage Utils/model" category despite being in lora.py
0 inputs · 1 parameter
Sage_MultiModelPicker
Pick a model_info entry by index from a provided list.
Sage_MultiplyConditioningStrength
Compact description unavailable for this node.
0 inputs · 1 parameter
Sage_MultiSelectorDoubleClip
Compact description unavailable for this node.
0 inputs · 6 parameters
Sage_MultiSelectorFlexibleClip
Compact description unavailable for this node.
0 inputs · 4 parameters
Sage_MultiSelectorQuadClip
Compact description unavailable for this node.
0 inputs · 7 parameters
Sage_MultiSelectorSingleClip
Compact description unavailable for this node.
0 inputs · 5 parameters
Sage_MultiSelectorTripleClip
Compact description unavailable for this node.
0 inputs · 6 parameters
Sage_NormalizeConditioningStrength
Compact description unavailable for this node.
Sage_NumberToStr
Compact description unavailable for this node.
0 inputs · 1 parameter
Sage_OllamaAdvancedOptions
Get advanced options for Ollama LLMs.
0 inputs · 9 parameters
Sage_OllamaLLMPromptText
Ollama LLM Prompt (Text) Sends a text prompt to an Ollama LLM and returns the response. - prompt (STRING): The text prompt to send to the model (multiline, default: detailed description prompt) - model (STRING): The Ollama model to use (from available installed models) - seed (INT): Seed for random number generation (default: 0, range: 0 to 2^32-1) - loadforseconds (FLOAT): Time in seconds to keep model loaded, -1 for indefinitely (default: 0.0, range: -1.0 to 3600.0) - response (STRING): The LLM's text response to the prompt Use to generate text completions or responses from an Ollama LLM in your workflow. Requires Ollama to be installed and models to be available. - EXPERIMENTAL: This node is experimental and may change in future versions - Requires Ollama to be installed and running - Model list is populated from available Ollama models - Seed ensures reproducible results when set to same value - loadforseconds controls how long model stays in memory after generation - Default prompt is optimized for detailed descriptions - Raises error if Ollama is not available or model not found - Supports all text-based Ollama models
0 inputs · 5 parameters
Sage_OllamaLLMPromptVision
Ollama LLM Prompt (Vision) Sends a vision prompt with image input to an Ollama LLM and returns the response. - prompt (STRING): The text prompt to send to the model (multiline, default: detailed image description prompt) - model (STRING): The Ollama vision model to use (from available installed vision models) - image (IMAGE): The image to analyze and describe - seed (INT): Seed for random number generation (default: 0, range: 0 to 2^32-1) - loadforseconds (FLOAT): Time in seconds to keep model loaded, -1 for indefinitely (default: 0.0, range: -1.0 to 3600.0) - response (STRING): The LLM's text response describing the image Use to generate vision-based completions or responses from an Ollama LLM in your workflow. Ideal for image captioning, analysis, and description tasks. - EXPERIMENTAL: This node is experimental and may change in future versions - Requires Ollama to be installed and running - Only shows vision-capable models in the dropdown - Image input is required for vision models - Seed ensures reproducible results when set to same value - loadforseconds controls how long model stays in memory after generation - Default prompt is optimized for detailed image descriptions suitable for AI generators - Raises error if Ollama is not available, model not found, or no image provided - Supports multimodal input (image + text prompt)
1 input · 5 parameters
Sage_OllamaLLMPromptVisionRefine
An advanced Ollama LLM node that provides refined vision-based prompt generation with two-stage processing. This node first generates an initial description using a vision model, then refines that description using a separate text model for enhanced quality and detail. - prompt (STRING): Initial prompt for the vision model (default: detailed image description prompt) - model (COMBO): Available Ollama vision model for initial analysis - image (IMAGE): The input image to analyze - seed (INT): Random seed for the initial vision model (0 to 2^32-1) - refineprompt (STRING): Prompt for refining the initial description (default: rewrite to be more vivid and detailed) - refinemodel (COMBO): Available Ollama text model for refinement - refineseed (INT): Random seed for the refinement model (0 to 2^32-1) - initialresponse (STRING): The original response from the vision model - refinedresponse (STRING): The enhanced response after refinement - Stage 1: Vision model analyzes the image using the initial prompt - Stage 2: Text model refines the vision output using the refine prompt - Dual seeding: Separate seeds for reproducible results in both stages - Model flexibility: Can use different models for vision and refinement tasks This node is perfect for: - Creating high-quality prompt descriptions from images with iterative improvement - Generating detailed image analysis that's then enhanced for clarity and engagement - Building sophisticated prompts through a two-model approach - Achieving consistent, reproducible results with separate seeding control 1. Vision model (e.g., llava:13b)…
1 input · 6 parameters
Sage_ParseMetadataFlexible
Reverse parser for A1111 Full format metadata string.
0 inputs · 8 parameters
Sage_PercentageToFloat
Convert a percentage value to a float between 0 and 1.
0 inputs · 1 parameter
Sage_PonyPrefix
Add Pony v6 Prefixes Creates Pony v6 prefixes for prompt engineering based on score, rating, and source parameters. - addscore (boolean, required): Whether to include score prefixes (score9, score8up, etc.) - rating (string, required): Content rating - Options: "none", "safe", "questionable", "explicit" - source (string, required): Content source - Options: "none", "pony", "furry", "anime", "cartoon", "3d", "western", "comic", "monster" - STRING: Generated prefix string containing the selected score, rating, and source tags Use to automatically generate appropriate prefixes for Pony v6 models. The node combines score tags (if enabled), source tags, and rating tags into a properly formatted prefix string that can be prepended to your main prompt.
0 inputs · 5 parameters
Sage_PonyRatingv6
Creates a rating string for pony prompts (v6 style).
0 inputs · 1 parameter
Sage_PonyRatingv7
Creates a rating string for pony prompts (v7 style).
0 inputs · 1 parameter
Sage_PonyScore
Creates a score string for pony prompts.
0 inputs · 3 parameters
Sage_PonySource
Creates a source string for pony prompts.
0 inputs · 1 parameter
Sage_PonyStyle
Add Pony Style Adds three-letter artist style codes from Pony v6 to prompts. - style (multi-select): Choose from hundreds of three-letter artist style codes - Multi-select dropdown with chip display - Options include: aav, aax, aba, aca, acb, acl, etc. (extensive list of artist style codes) - text (STRING): Comma-separated list of selected style codes Use to add specific artist styles to your Pony v6 prompts. Select one or more three-letter codes that correspond to different artistic styles. The node will output them as a comma-separated string that can be added to your prompt. - Contains hundreds of three-letter artist style codes specific to Pony v6 - Supports multi-selection for combining multiple artistic influences - Output is formatted as comma-separated values for easy prompt integration - Style codes are based on the Pony v6 model's training data and recognized artistic styles
Sage_PonyStyleCluster
Creates a style cluster string for pony prompts.
0 inputs · 1 parameter
Sage_PromptText
Compact description unavailable for this node.
Sage_QuadCLIPSelector
Compact description unavailable for this node.
0 inputs · 4 parameters
Sage_QuickLoraStack
Simplified lora stack node without clip_weight.
0 inputs · 3 parameters
Sage_QuickResPicker
A convenient node for quickly selecting common image resolutions from predefined aspect ratios with orientation and scaling options. - aspectratio (COMBO): Predefined aspect ratio selection - "1:1", "5:12", "9:16", "10:16", "5:7", "2:3", "3:4", "4:7", "7:9", "8:10", "13:19" - orientation (COMBO): "Portrait" or "Landscape" orientation - multiplier (FLOAT): Scale factor for the resolution (default: 1.0, range: 0.1-10.0, step: 0.1) - width (INT): The calculated width dimension - height (INT): The calculated height dimension Pick a resolution from a list of common aspect ratios. The multiplier can be used to scale the resolution up or down, rounded to the nearest unit of 64. - 1:1 - Square (1024 x 1024) - 5:12 - Portrait (512 x 1216) - 9:16 - Portrait (720 x 1280) - 10:16 - Portrait (640 x 1024) - 5:7 - Portrait (1280 x 1792) - 2:3 - Portrait (768 x 1152) - 3:4 - Portrait (768 x 1024) - 4:7 - Portrait (768 x 1344) - 7:9 - Portrait (896 x 1152) - 8:10 - Portrait (1024 x 1280) - 13:19 - Portrait (832 x 1216) - All base resolutions are listed in portrait orientation - Landscape orientation swaps width and height - Multiplier allows scaling while maintaining aspect ratio - All outputs are rounded to nearest multiple of 64 for optimal AI generation - Defaults to 1:1 aspect ratio if invalid selection - Useful for quickly setting up standard image dimensions - Covers common aspect ratios used in AI image generation
0 inputs · 3 parameters
Sage_ReferenceImage
Compact description unavailable for this node.
3 inputs · 0 parameters
Sage_SamplerInfo
Sampler Info Collects and packages sampler settings for use in metadata construction and KSampler nodes. - seed (INT, required): Random seed for noise generation - Default: 0, Range: 0 to 2^64-1 - steps (INT, required): Number of denoising steps - Default: 20, Range: 1 to 10,000 - cfg (FLOAT, required): Classifier-Free Guidance scale - Default: 5.5, Range: 0.0 to 100.0, Step: 0.1 - samplername (dropdown, required): Sampling algorithm - Default: "dpmpp2m" - Options: All available ComfyUI samplers - scheduler (dropdown, required): Noise scheduling algorithm - Default: "beta" - Options: All available ComfyUI schedulers - SAMPLERINFO: Dictionary containing all sampler parameters Connect to both the Construct Metadata node and KSampler nodes to standardize and share sampler configuration across your workflow. This node centralizes sampler settings, making it easy to maintain consistency and modify parameters in one place. - Designed to work with SageKSampler and other sampler nodes - Output should be routed to both metadata construction and sampling nodes - CFG scale balances creativity and prompt adherence - Higher CFG values increase prompt adherence but may reduce quality if too high - Sampler algorithm affects quality, speed, and style of generation - Scheduler controls how noise is gradually removed during denoising - Centralizes sampler configuration for workflow consistency
0 inputs · 5 parameters
Sage_SamplerInfoNoCFG
Grabs most of the sampler info.
0 inputs · 4 parameters
Sage_SamplerSelector
Selects a sampler for use in the pipeline.
0 inputs · 1 parameter
Sage_SaveImageWithMetadata
Save Image w/ Added Metadata Saves images with comprehensive metadata options including custom parameters, extra metadata, and ComfyUI workflow information. - images (IMAGE): The images to save - filenameprefix (STRING): The prefix for the file to save, supports formatting like %date:yyyy-MM-dd% or %Empty Latent Image.width% (default: "ComfyUIMeta") - includenodemetadata (BOOLEAN): Whether to include ComfyUI prompt/workflow metadata (default: True) - includeextrapnginfometadata (BOOLEAN): Whether to include extra PNG info metadata (default: True) - parammetadata (STRING): Metadata string for "parameters" field (A1111-style) - extrametadata (STRING): Additional metadata for "Extra" field - prompt: ComfyUI workflow prompt (automatically provided) - extrapnginfo: Extra PNG info (automatically provided) - None (Output Node): Saves images to ComfyUI output directory Use to save images with custom and standard metadata for better tracking and sharing. Connect parammetadata from Construct Metadata node for A1111-compatible metadata. - Saves images as PNG with customizable compression level - parammetadata is stored under "parameters" key (A1111 standard) - extrametadata is stored under "Extra" key - Filename prefix supports dynamic formatting with node values and dates - Batch number support with %batchnum% placeholder - Respects ComfyUI's --disable-metadata flag - Compatible with A1111 and Civitai metadata standards - Output node that appears in ComfyUI's image output interface - Useful for workflows requiring comprehensive metadata tracking
1 input · 6 parameters
Sage_SaveText
A utility node for saving text content to a file. This node allows you to write text strings to files on disk with customizable filename and extension. - filenameprefix (STRING): The prefix for the file to save (default: "ComfyUIText"). This may include formatting information such as %date:yyyy-MM-dd% to include values from nodes. - fileextension (STRING): The file extension to use for the saved file (default: "txt") - text (STRING): The text content to save to file - filepath (STRING): The full path of the saved file This node is helpful for: - Saving generated prompts or text to files - Creating logs of your workflow outputs - Storing intermediate text results for later use - Debugging by writing node outputs to files - The filename prefix supports formatting patterns for dynamic naming - File extension can be specified with or without the leading dot - A counter will be added to the end of the filename, based on how many files with that name there are. - If the file path is invalid, an error will be raised
0 inputs · 4 parameters
Sage_SchedulerSelector
Selects a scheduler for use in the pipeline.
0 inputs · 2 parameters
Sage_SetText
Set Text Sets and outputs a text string with optional prefix and suffix. Useful for providing static or user-defined text to workflows. - str (STRING): The main text string to output - prefix (STRING): Text to add before the main string - suffix (STRING): Text to add after the main string - str (STRING): The combined text string (prefix + str + suffix) Connect to nodes that require a text input, such as prompt or metadata nodes. The node combines prefix, main text, and suffix into a single output string.
0 inputs · 3 parameters
Sage_SetTextWithNum
Compact description unavailable for this node.
0 inputs · 4 parameters
Sage_SetTextWithoutComments
Compact description unavailable for this node.
0 inputs · 3 parameters
Sage_SingleCLIPTextEncode
Compact description unavailable for this node.
Sage_SingleCLIPTextImageEncode
Compact description unavailable for this node.
Sage_StackLoraStack
Compact description unavailable for this node.
Sage_StylePromptFromConfig
Compact description unavailable for this node.
0 inputs · 3 parameters
Sage_SystemPrompt
Compact description unavailable for this node.
Sage_TextRandomLine
Compact description unavailable for this node.
0 inputs · 2 parameters
Sage_TextSelectLine
Compact description unavailable for this node.
0 inputs · 2 parameters
Sage_TextSubstitution
Compact description unavailable for this node.
Verified Sep 12
Verified Sep 12