ComfyUI-GU_Nodepack Introduction
ComfyUI-GU_Nodepack is a collection of ComfyUI nodes for organizing creative workflows. Its tools help you build output filenames, keep prompts and generation details with your images, work with LoRAs, choose resolutions and seeds, and handle common video and text tasks.
For example, you can take the name of an image you loaded, add a project name and date, and use the resulting path to save both an output image and a matching text file of generation details. You can also use the pack to try random LoRA combinations, label an image with its settings, or keep a list of seeds worth revisiting.
The author tested the nodepack on Windows. Linux and macOS support is best-effort. It is designed for a single-user ComfyUI setup; some functions may have race conditions on a multi-user server.
How ComfyUI-GU_Nodepack Works
In ComfyUI, each node performs one part of a workflow. ComfyUI-GU_Nodepack adds nodes that pass along a model, an image, or a video-related value, as well as nodes that produce useful information about a generation. You connect their outputs to other nodes just as you would elsewhere in ComfyUI.
Think of the pack as a set of small studio assistants:
- A naming assistant can read an input filename, combine it with project details, and prepare a consistent output path.
- A record keeper can collect prompts, settings, hardware details, elapsed time, and seeds so you can identify or revisit an image later.
- A workflow helper can select a sampler by number, choose one of several inputs, split text into lines, or calculate an appropriate video frame count.
Some nodes intentionally run every time you queue a workflow, even when their inputs appear unchanged. This matters for tasks such as refreshing a filename, recording seeds, choosing random LoRAs, or measuring time. Individual nodes that behave this way are identified below.
ComfyUI-GU_Nodepack Features
Video and frame planning
Calculate Frame Count reports the effective number of frames when loading a video with frame limits or frame skipping. Supply the video's total frame count (frames_total), the requested number of frames (load_frame), and the interval for loading every nth frame (load_nth). The integer result helps you understand how much material the rest of the workflow will process.
Set Frames Video sets a frame count that follows the N*8 + 1 pattern required by the LTX- and WAN-style pipelines described in the original documentation. Use it when a video workflow needs a valid count in that form.
Text and prompt tools
Concat Strings joins a configurable number of text inputs. Set inputs, then click Update inputs to expose the fields you need. The delim setting places a separator between pieces; its default, _, is useful for filenames. Turn on isfile when the result will be used as a filename: forbidden characters are replaced with underscores.
Extract Prompt looks through image metadata and gathers text from nodes whose titles or class names suggest prompts, text encoding, or positive and negative conditioning. Connect metadata from an upstream loader, such as Load Image with Metadata, to metadata_raw. The node returns pos_prompt, neg_prompt, and unsure_prompt. Review the results before reusing them: the labels indicate the node's best classification, and text it cannot classify confidently goes into unsure_prompt.
String Lines turns multiline text into a list, ignoring empty lines. It returns the non-empty str_list, the line chosen by sel_index as str_selected, and lines_count. This is useful when you have one prompt idea or label per line and want to select one within a workflow.
String to Values splits text using a delimiter and returns elements in the type selected by out_type. Use it when one text field contains several values that a downstream node needs separately.
Text Edit gives you an editable field for incoming, dynamically generated text, such as an image-to-text description. Queue the workflow once to populate the field, make your changes, then queue it again to send the edited text downstream. Normally, each run resets the field to the incoming text; enable protect to block that refresh while you work. The edited text is not preserved across workflow or ComfyUI restarts, and the author warns that only one Text Edit node per graph can currently be used reliably. This node runs on every queue.
Identifying inputs and choosing workflow values
Get Media Input Name retrieves the filename of an upstream image, video, or audio loader. It handles reroutes and, by default, removes the file extension through strip_ext. Its output is useful as the starting point for a new output filename. It runs on every queue; the author notes possible ordering issues when several copies are present in one graph.
Get Model Filename retrieves an upstream model loader's filename, such as a .safetensor or .ckpt name. It handles reroutes and runs on every queue. You can use the result to identify the model used for an output.
Get Node Active reports whether an upstream node is active. This lets another part of a workflow respond to that node's state.
Get Sampler Indexed and Get Scheduler Indexed select a sampler or scheduler by its position in ComfyUI's corresponding system list. They are convenient when a workflow cycles through choices by index. Each returns the selected value, its name, and total, which indicates the list length.
Get Sampler Indexed node Get Scheduler Indexed node
Switch Any by Index passes through one of its inputs, whatever that input's data type. Set inputs to a value from 2 to 10 and click Update inputs; then use index to choose an input, numbered from 1. An index outside the available range selects the nearest end input instead. This is useful for choosing between alternative branches without changing their data.
Resolution Select provides a range of popular image and video resolutions, including Midjourney resolutions. Turn on portrait to swap the output width and height. The swap also applies to manually entered dimensions, so check the resulting orientation before a long run.
Sides Pack combines four integer values into one tuple for a downstream node. Sides Unpack separates that tuple back into its four values. Together, they make per-side settings, such as outpainting padding, easier to carry through a workflow.
Sides Pack node Sides Unpack node
LoRA tools
Lora Randomizer chooses LoRAs and weights at random from the folder named in lora_subfolder. It scans that folder only, not its child folders. Set loras_max to limit the selection to at most six LoRAs; if fewer are available, it uses as many as it finds. strength_min and strength_max set the possible model-strength range, while clip_min and clip_max do the same for CLIP strength. Although the accepted range is -100 to 100, the original documentation notes that values around 0 to 1 are typical. Smaller ranges give you more controlled experiments; wider ranges allow more variation.
You can supply pos_prompt and use append_tags to include selected LoRA tags in ready_prompt. Alongside the modified model and CLIP outputs, the node returns lora_tags and a lora_list of application settings in the form <lora:name:str:clp>. It runs on every queue, making it useful for repeated variation.
Lora Stack Extended is the more deliberate option: configure up to six individual LoRAs, each with its own model and CLIP strengths. Each slot's enable checkbox lets you compare a result with and without that LoRA without losing its settings. Like the randomizer, it can append tags to a supplied positive prompt and returns ready_prompt, lora_tags, and lora_list, as well as the modified model and CLIP.
Motion Lora Stack groups animation LoRAs used by AnimateDiff from the __animatediff_motion_lora__ folder in the ComfyUI tree. It also returns the names of LoRAs used, which you can include in an information file or video filename.
Turbo Lora optionally attaches a selected accelerating LoRA to the incoming model. Choose its file with turbo_lora, switch it on or off with enable, and set separate steps_normal and steps_turbo counts. Its steps output follows the selected mode, so a connected workflow can change step count along with the LoRA. It also returns a lora_list parameter string.
Project names, output paths, and records
Project Name supplies a project code. Enter one in project or choose a preset from actual_proj_codes. The preset codes are stored in web/proj_codes.txt. Use its output to keep related images and records together.
Project Path builds a base path and filename for an image or video saved inside ComfyUI's output folder. You can add a project, an upscale folder, a date folder, a descriptive suffix, a model-type code, a process type, technical details, and a connected seed. Only supplied or enabled parts are included. ComfyUI appends the final sequence number and file extension separately.
A practical way to set it up is:
- Set
projectto group work under(work)/<project>/. When a project is set, slashes in its name and in the filename are replaced with_instead of creating extra folders. - Choose a source name: enable
use_custom_nameand entercustom_name, or connect upstream text tofilename, perhaps from Get Media Input Name. Leaveprojectempty if you need a subfolder specified throughcustom_name; the(work)folder is then not used. - Enable
use_datefor aYYYY_MM_DD/folder. Enableis_upscalefor an_UP/folder and, whenup_factoris at least 2, an_upNxfilename addition; the factor is rounded up to an integer. - Use
change_fromandchange_toto replace part of the source name, andname_suffixfor a further description. - Add optional
model_type,proc_type,tech_suffix, and a connectedseedto make outputs easier to identify. Model-type codes are customizable inweb/model_codes.txt. Image Label can supply a technical suffix.
The proc_type field can describe processes such as text-to-image, image-to-image, or inpainting. The author recommends omitting it for upscales, which already receive their own folder and filename treatment. A minimal final name might be test_00001_.png; a more descriptive one can include a project, date, upscale factor, model code, and technical suffix.
Image Label assembles readable text describing a generation and can add caption blocks to the bottom of an image when you also supply the image. You can configure the caption font face and size; its height adjusts to the amount of text. The outputs separate information into tech_label for hardware details, pos_label and neg_label for prompts, gen_label for generation settings and elapsed time, and tech_suffix for a short indication of features such as ControlNet, DetailDaemon, or Redux. You can pass tech_suffix to Project Path to reflect those details in a filename. For elapsed time, connect time_ctrl to the result of a TimerON / TimerOFF pair.
Save Image Info writes a text file using file_name and can include the information blocks from Image Label. Feeding it the base path from Project Path is recommended when you want the text record beside the corresponding image with a matching name. Subfolders within ComfyUI's output directory are supported. The node runs on every queue.
Seed List generates seed values and returns accumulated, non-fixed values in prev_seeds. Enable store_to_file to write them line by line to a text file. Use a distinct filename for each task so separate Seed List nodes do not overwrite the same default seedfile.txt. reset_list clears the accumulated list on the next run, except when init_from_file is enabled. With init_from_file, the node tries to load numbers from filename; if the file is missing or the name is empty, it starts from scratch.
The optional project input can place the seed file alongside project output. If you enter a value there manually, provide a final file path, not just a folder. The node runs on every queue.
TimerON starts a counter, and TimerOFF stops it. TimerOFF returns both a readable hh:mm:ss value and time_raw in whole seconds. Place the pair around the work you want to measure, then pass the readable result to Image Label if you want the duration recorded with an image. TimerON runs on every queue.
File-path safety
Paths used by Save Image Info and Seed List are confined to ComfyUI's output directory. Subfolders inside that directory are allowed. A path that tries to leave it, including one with .., an absolute path, or a drive prefix such as C:, is rejected with a console warning; the node does not read or write that file. Other unsafe filename characters are sanitized. This means project organization remains available without allowing a workflow-provided path to write elsewhere on the machine.
Troubleshooting ComfyUI-GU_Nodepack
Extra inputs such as str_-2, str_-1, str_0, any_-1, or any_0 appear. What should I do?
The author notes that older ComfyUI versions may show these unintended inputs on Concat Strings or Switch Any by Index. Do not connect anything to them. Use the normally numbered inputs and click Update inputs after changing the input count.
Text Edit did not send my changes on the first run.
Its first run fills the editable field from incoming dynamic text. Edit that field, then queue the workflow a second time to forward its contents. Enable protect if later runs would otherwise replace your edit with fresh incoming text. Do not rely on the field to retain edits after restarting the workflow or ComfyUI.
Text Edit behaves unexpectedly when I use several copies.
The author warns that only one Text Edit node per graph can currently be used reliably. Try a graph with a single Text Edit node to see whether the issue persists.
A filename or seed file was not written.
Check the ComfyUI console for a path warning. Make sure the requested path stays within ComfyUI's output folder, contains no .. component, and is not an absolute path or drive-prefixed path. For Seed List, also check that a manually entered project value names a file rather than only a folder. When using several Seed List nodes, give each task a distinct filename.
Lora Randomizer did not find a LoRA in a nested folder.
Point lora_subfolder at the folder that directly contains the LoRA files. The node does not search child folders. It can select no more than six LoRAs, even if the folder contains more.
The selected resolution has the wrong orientation.
Check portrait in Resolution Select. It swaps width and height for both presets and manual values.
My output filename is missing a detail I expected.
Check that the corresponding Project Path input is populated or its option is enabled. For example, the date requires use_date, a custom name requires use_custom_name, and a seed is added when its input is connected. The final sequence number and file extension are appended by ComfyUI, not by Project Path.
Several Get Media Input Name nodes return results in an unexpected order.
The author notes a possible ordering issue when multiple copies appear in one graph. Check which loader each node is connected to, including connections through reroutes, and simplify the affected graph while diagnosing it.
Does this nodepack include its own image-generation models?
The documentation describes tools for using or identifying models and LoRAs already present in a ComfyUI workflow. It does not identify a separate set of image-generation models supplied with the nodepack.
Learn More about ComfyUI-GU_Nodepack
The example folder contains an example workflow and a test image for checking most of the nodes. It is a useful place to see how the tools connect before adding them to a larger creative workflow.
For project naming presets and model-type labels used by Project Path, see web/proj_codes.txt and web/model_codes.txt. The node images throughout this README show the available controls and connections for each feature.
