Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > comfyui-jjmaden-random-nodes

ComfyUI Extension: comfyui-jjmaden-random-nodes

Repo Name

comfyui-jjmaden-random-nodes

Author
jjmaden (Account age: 0 days)
Nodes
View all nodes(1)
Latest Updated
2026-11-03
Github Stars
0.0K

How to Install comfyui-jjmaden-random-nodes

Install this extension via the ComfyUI Manager by searching for comfyui-jjmaden-random-nodes
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter comfyui-jjmaden-random-nodes in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

comfyui-jjmaden-random-nodes Description

comfyui-jjmaden-random-nodes adds 12 ComfyUI nodes for image routing, cropping, prompt variation, LoRA control, video resizing, and MiniMax H3 reference media, audio-video, and chat workflows.

comfyui-jjmaden-random-nodes Introduction

comfyui-jjmaden-random-nodes is a collection of 12 custom nodes for ComfyUI. Seven are general workflow tools for choosing images, switching between branches, resizing video inputs, managing LoRAs, and varying prompts. Five support MiniMax H3 video workflows by providing ways to prepare media, build prompts, and work with combined audio-and-video data.

You can use the utility nodes independently. For example, Image Hub helps you pick a reference image from a folder, while Random Line From File can select a different prompt idea from a text list. The MiniMax H3 nodes are more specialized: they help assemble workflows that use keyframes, reference pictures, video, and audio together.

Important limitation: The five MiniMax H3 nodes were built to fill gaps in a privately distributed workflow. Some behavior is inferred from that workflow rather than verified against its original nodes. Read the model card and check your renders before relying on their output for important work.

How comfyui-jjmaden-random-nodes Works

A ComfyUI workflow is a chain of connected nodes. Each node receives something, such as an image, a prompt, or model settings, and passes a result to the next node. This collection adds a few useful ways to control that chain:

  • Choose a route: Switch nodes let one workflow contain several alternatives while you select the one to use for a run.
  • Prepare an input: Image Hub selects and crops an image; LTX Latent Resizer adjusts image dimensions and, optionally, the number of video frames.
  • Vary a prompt: Random Line From File picks a line from your text file. You decide whether that choice follows the image seed or uses its own randomness.
  • Combine references: The MiniMax H3 nodes organize pictures, video, and audio into inputs for a video workflow.

Think of a switch node as a junction in a studio setup: you can leave several arrangements connected and choose which one feeds the next step. Some inputs are lazy, meaning ComfyUI does not need to run their upstream branch unless that branch is selected. The exact behavior differs by node; the Sampler Switch, for instance, always evaluates Set 1 because it also serves as its fallback.

comfyui-jjmaden-random-nodes Features

Dynamic Image Router

The Dynamic Image Router sends one incoming image batch to one of ten outputs. Set select_path to a number from 1 to 10 to choose the destination.

On ComfyUI builds that support execution blocking, the other nine branches do not run. This is useful when you have several possible downstream treatments for an image, such as different previews or processing paths, but only want to execute one. On older builds, the unselected outputs receive an empty image batch instead; a downstream node that assumes an image is always present may still report an error.

Image Hub

Image Hub lets you browse and load an image, along with a matching mask, from ComfyUI’s input or output folder or from a custom folder. Its searchable filename list includes supported images in subfolders, so you can choose a file without repeatedly typing its path.

You can work directly from the node:

  • Refresh File List rescans the selected folder when a new file is not showing.
  • Paste from Clipboard uploads a copied image to ComfyUI’s input folder and selects it. Your browser must permit clipboard access, and the page must be on localhost or HTTPS.
  • Thumbnail preview shows the selected image and its original dimensions. The thumbnail is for display; the node loads the actual image file.
  • Interactive crop lets you draw, move, resize, or clear a rectangle on the preview. The crop is applied to the full-resolution image and is saved with the workflow.

For output size, max_megapixels sets a ceiling and only scales oversized images down. Setting both force_width and force_height instead produces an exact size, cropping from the center first if needed. Exact width and height take priority over the megapixel limit.

If an image has transparency, Image Hub derives its mask from that transparency. Otherwise, it produces an empty mask at the image’s dimensions. You can use it to bring a copied reference image into a workflow quickly or browse images you have already generated.

Model Switch (DualModelDPDTSwitch)

The Model Switch chooses between two pairs of model inputs, labeled Set A and Set B, and outputs the selected high and low models. One use is keeping text-to-video and image-to-video model pairs connected in the same workflow.

Only the selected pair’s upstream model loaders are evaluated. This can avoid loading the unused pair for that run. Both inputs in the chosen pair must be connected; if one is missing, the node identifies it in an error.

The switch does not provide model files. It routes models that your workflow loads elsewhere.

Sampler Switch (MultiInputSamplerSwitch)

The Sampler Switch chooses one of up to five sets of positive conditioning, negative conditioning, and latent input for a sampler. Set select_set from 1 to 5 to try different prompt-and-latent combinations through the same downstream sampling chain.

Set 1 is required and always runs because it is the fallback. Sets 2–5 are optional and lazy. If part of a selected optional set is missing, the node uses that part from Set 1 and prints a warning. For a clean comparison between complete alternatives, connect all three inputs in every set you intend to test.

LTX Latent Resizer

Despite its name, LTX Latent Resizer works on images before they become latents. Place it before a VAE Encode step when you need image dimensions, or a video frame count, that suit your target workflow.

scale_factor changes the size, and multiple_of rounds the resulting width and height up to a required multiple; the default is 32. You can choose bilinear, bicubic, nearest, or area resizing.

The optional video_preset also trims frames from the end of an image batch to meet a model family’s frame-count pattern:

PresetValid frame-count pattern
LTX-Video8k + 1
Hunyuan Video4k + 1
WAN 2.1 / 2.24k + 1
CustomYour chosen multiple and remainder
NoneLeaves the frame count unchanged

The node never adds or duplicates frames. If you need a particular length, supply at least that many frames before trimming. It outputs the resized image, final width, final height, and final frame count.

Power Lora Hooks

Power Lora Hooks combines up to eight LoRAs into a reusable HOOKS output. Each slot has an on/off setting, a LoRA selection, and separate model and CLIP strengths. A disabled slot, a slot set to None, or a slot with both strengths at zero contributes nothing.

You can connect positive or negative conditioning to receive the hooks, or pass the HOOKS output to another hook-compatible node. This makes it possible, for example, to apply a style LoRA to positive conditioning without applying that same stack to negative conditioning. You can also connect multiple Power Lora Hooks nodes if eight slots are not enough.

This node uses ComfyUI’s experimental hooks system and does not directly patch the model and CLIP the way a conventional LoRA loader does. Check compatibility in the workflow you intend to use.

Random Line From File (Wildcard)

Random Line From File chooses one non-empty, non-comment line from a plain-text file. You can use the lines as alternate subjects, styles, scene descriptions, or other prompt fragments. prefix_text adds text before the chosen line, and lora_trigger replaces a literal [trigger] token when it appears in a line.

Its prompt_mode controls the choice:

  • Follow KSampler/Seed Field: Tie the line choice to the supplied seed.
  • Use Dedicated Prompt Seed: Use prompt_fixed_seed to control prompt variation separately.
  • True Independent Random: Make a fresh random choice independently of the supplied seed.

The node also passes its seed input through as seed_output. Blank lines and lines beginning with # are ignored. Save the file as UTF-8; a missing file, an unreadable encoding, or a file with no usable lines produces a clear error rather than sending an error message into your prompt.

MiniMax H3 Resolution Selector

The MiniMax H3 Resolution Selector produces a width, height, size label, and frame count for a MiniMax H3 generation. Choose an aspect ratio and resolution tier instead of calculating dimensions by hand. Its presets cover square, portrait, landscape, and wider formats; duration and fps determine a frame count aligned to MiniMax H3’s required grid.

You can override the preset by setting both custom_width and custom_height. Alternatively, connect reference_image to use its aspect ratio while staying within the selected resolution’s pixel budget. Custom dimensions take priority over the reference-image sizing.

Changing aspect_ratio filters the resolution choices in the node’s interface. A saved workflow is not silently given a different resolution merely because you reopened it.

MiniMax H3 Concat AV Latent

MiniMax H3 Concat AV Latent combines a separately prepared video latent and audio latent into the combined structure expected by MiniMax H3 audio-and-video workflows. This is useful when you want to encode the picture and sound separately before bringing them together.

The expected inputs are a 24-channel video latent and a 32-channel audio latent. If an input does not have the expected shape, the node reports which one needs attention.

MiniMax H3 Unified To Video

MiniMax H3 Unified To Video prepares conditioning and an empty combined audio-and-video latent for generation. It can draw on a text prompt, first and last keyframes, reference pictures, a reference video and its audio, and standalone audio.

Choose a mode to decide which media pathways participate:

ModeWhat it uses
keyframeFirst and last frames only
referenceReference media, not keyframes
hybrid or autoConnected keyframes and references
textPrompt text without media

For reference images, ref_image_size: match uses the generation’s pixel budget, while max uses a larger reference-image pipeline that may take more resources. The node’s report and media_map_json outputs help you inspect which media was attached, its tag order, and warnings. For example, they report when a reference video is trimmed because it exceeds the requested output length.

You can connect individual reference sockets or an H3_REFS bundle from the separately installed MiniMaxH3MediaLoader or MiniMaxH3PromptBuilder. A directly connected socket takes priority over the matching item in the bundle.

Compatibility caution: This node’s approach to combining and tagging different media types is partly inferred from a saved workflow. Its results are not verified as identical to the unavailable private node.

MiniMax H3 Audio Lock

MiniMax H3 Audio Lock encodes a reference audio clip and blends it into the audio portion of an existing MiniMax H3 audio-and-video latent. It requires a combined AV latent, an audio VAE, and audio supplied either directly through audio_1 or through a references bundle.

strength controls the blend: 0 leaves the target audio unchanged, while 1 replaces it with the encoded reference. mode: lock loops the reference across the target duration; mode: blend affects the beginning and leaves the remainder of the target audio in place.

This behavior is an unverified interpretation of the unavailable original node. Treat the report’s disclaimer seriously and listen to the rendered result. See the model card for its limitations.

MiniMax H3 Multimodal Chat

MiniMax H3 Multimodal Chat helps you draft and revise prompts through a chat conversation. Its prompt_text output extracts a finished prompt from the reply, ready to connect to another node. Feed its chat_history output back into the corresponding input when you want the next turn to remember the conversation; set new_chat when you want to start over.

It supports local backends (Ollama, LM Studio, and KoboldCpp), cloud backends (OpenAI, Anthropic, and Gemini), or generation through a compatible clip already loaded in ComfyUI. You supply the model used by the chosen backend; this extension does not include a chat model. The comfyui_clip option requires a generation-capable model, not an ordinary CLIP or T5 text encoder, and its current implementation is text-only.

For supported chat backends, the node can send a limited selection of reference images to help describe a scene: up to two pictures and one video frame. It does not send audio content; it only tells the chat model that audio references are attached. Avoid assuming its reply reflects sounds the model has heard. If a backend cannot be reached, the node returns an error reply and report instead of stopping the queue.

The built-in prompt guidance covers spoken dialogue, including stable speaker labels and exact quoted lines. If you replaced that guidance with an older custom system_prompt, review it when dialogue is important to your video.

API-key privacy: For cloud backends, the node can use the matching OPENAI_API_KEY, ANTHROPIC_API_KEY, or GEMINI_API_KEY environment variable. A key typed into a node field is saved as plain text in the workflow JSON, so take care before sharing that workflow.

What's New with comfyui-jjmaden-random-nodes

The supplied README describes changes but does not provide version numbers or a dated release history. These are the documented improvements, without assigning them to a particular release:

  • Image Hub gained a searchable filename dropdown, file-list refresh, clipboard paste, thumbnail preview, interactive crop, sorting controls, and output-sizing options. Existing saved filenames should still work, but refresh the list before choosing a newly added file.
  • Random Line From File now isolates its randomness from other nodes, skips comment lines, detects edits to its text file, caches parsed lines, and gives clearer errors for missing, empty, or incorrectly encoded files.
  • LTX Latent Resizer added optional video frame-count presets while retaining spatial-only resizing as its default behavior.
  • MiniMax H3 Unified To Video added direct use of an H3_REFS bundle and reports reference-video trimming. Its individual reference sockets were renamed to match the Reference Splitter’s numbering.
  • MiniMax H3 Audio Lock added clearer validation for missing audio and support for an H3_REFS bundle. Its former audio socket is now audio_1.
  • MiniMax H3 Multimodal Chat supports several local and cloud backends as well as compatible generation-capable ComfyUI CLIP models. It also includes reference-media handling, chat-history output, dialogue guidance, and backend-dependent unloading controls.

For older saved MiniMax H3 workflows: If a node still uses the former ref_image_0-style sockets, or Audio Lock uses the former audio socket, replace that node instance with a fresh one from the node menu and reconnect its inputs. Reloading the old instance alone may not update its sockets correctly.

Troubleshooting comfyui-jjmaden-random-nodes

My new image does not appear in Image Hub

  1. Check that folder_type points to the folder containing the image. For custom, check custom_path too.
  2. Click Refresh File List to rescan without restarting ComfyUI.
  3. Open the filename dropdown and type part of the name to filter it.

If the folder has no recognized images, the list may show other files as a fallback. Selecting a non-image file will produce an error naming that file.

Paste from Clipboard does not work

  1. Check that an image, rather than just a file path or text, is on your clipboard.
  2. Open ComfyUI through localhost or HTTPS.
  3. Allow clipboard access when your browser asks.

A switched branch behaves unexpectedly

  1. Confirm the selector points to the intended output or input set.
  2. For Model Switch, connect both the high and low model inputs in the selected pair.
  3. For Sampler Switch, connect all three inputs in an optional set if you do not want missing pieces to fall back to Set 1.
  4. If an unselected Dynamic Image Router branch errors on an older ComfyUI build, check whether its downstream node assumes an image batch cannot be empty.

Random Line From File reports a file or encoding error

  1. Verify that text_path points to an existing text file.
  2. Make sure it contains at least one non-empty line that does not begin with #.
  3. Save the file as UTF-8, then queue the workflow again.

A MiniMax H3 node reports missing or incompatible inputs

  1. Check that your ComfyUI build supports the MiniMax H3 structures needed by the workflow.
  2. For Concat AV Latent, check that the video and audio latents have the expected 24 and 32 channels.
  3. For Audio Lock, connect audio to audio_1 or provide a bundle containing standalone audio. An empty audio socket is not a usable reference.
  4. If an older workflow shows obsolete socket names, place a fresh node and rewire it.

For limitations specific to the reconstructed MiniMax H3 nodes, consult the model card.

Multimodal Chat cannot produce a useful reply

  1. Check that the chosen local backend is running, or that the cloud backend has its required API key.
  2. Confirm that base_url and model refer to the backend and model you intend to use.
  3. If using comfyui_clip, connect a generation-capable CLIP. Ordinary contrastive CLIP and T5 encoders cannot generate chat text through this option.
  4. Read the node’s report output for the specific error or limitation.

Frequently asked questions

Does the extension include image, video, or chat models?
No. Its nodes route or prepare models and media that you provide through ComfyUI or a chosen chat backend.

Will Image Hub crop the thumbnail instead of my original image?
No. You draw on the preview, but the saved crop coordinates are applied when the full-resolution file is loaded.

Does the video resizer add frames to reach a valid count?
No. Its video presets only trim frames from the end.

Can Multimodal Chat listen to an audio reference?
No. Audio is noted in the message sent to the chat model, but the audio content itself is not transmitted.

Are the reconstructed MiniMax H3 nodes guaranteed to match the private workflow’s original nodes?
No. Some behavior is inferred. Inspect the reports and rendered results, particularly when using Unified To Video or Audio Lock.

Learn More about comfyui-jjmaden-random-nodes

  • MiniMax H3 model card: Limitations and verification status for the reconstructed MiniMax H3 nodes.
  • ComfyUI-KJNodes: Source of MiniMaxH3MemoryEfficientSageAttentionPatch, a separate node that may appear in related workflows.
  • ComfyUI-ResolutionSelector: A separate selector whose reference-image sizing and duration-to-frame-count ideas informed this collection’s MiniMax H3 Resolution Selector.
  • rgthree-comfy: Includes the Power Lora Loader that inspired the compact, multi-LoRA layout of Power Lora Hooks.
  • ComfyUI-noEmbryo: Includes an image-loading node whose preview crop interaction inspired Image Hub’s approach.

Related MiniMax H3 workflows may also name nodes not included in this collection: MiniMaxH3SigmaShift belongs to ComfyUI core, MiniMaxH3MemoryEfficientSageAttentionPatch belongs to ComfyUI-KJNodes, and MiniMaxH3ReferenceSplitter belongs to the separate “Fantastic H3 Prompt Builder” pack.

comfyui-jjmaden-random-nodes Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

comfyui-jjmaden-random-nodes detailed guide | ComfyUI