Macro Nodes

Macro Nodes

Quality-of-life nodes for ComfyUI. Small, no extra dependencies, and each one solves a thing that was annoying enough to fix properly.

• Save Image (Macro) - save named after the model, with a numbered workflow .json and a per-node metadata switch
• Model Name Extractor - the model's name as a STRING, for any save node
• Prompt Prefix (Macro) - system prompts as .txt files, picked from a dropdown
• Concat Text (Macro) - join up to ten inputs of any type, with a delimiter that can hold newlines

Registry: macronodes · Publisher: the-macroman · MIT

Install

ComfyUI Manager - search for Macro Nodes.

CLI

comfy node install macronodes

Manually

cd ComfyUI/custom_nodes
git clone https://github.com/the-Macro-Man/ComfyUI-MacroNodes

Restart ComfyUI. There is nothing to pip install - the pack uses the standard library plus numpy and PIL, which ComfyUI already requires.

---

Nodes

Save Image (Macro)

SaveImage, named after the model that made the picture, with control over what goes with it.

Found under MacroNodes/image. Connect an IMAGE and that is it - the model name is read from the graph, so there is nothing to wire up and nothing to type.

| Input | Default | What it does | |---|---|---| | images | - | The images to save | | folder_prefix | "" | Prefixes the FOLDER, e.g. art_ → art_krea2_turbo/ | | file_prefix | "" | Prefixes the FILE, e.g. draft_ | | create_folder | on | Save into a subfolder named after the model | | embed_workflow | on | Off writes a clean PNG with no prompt or workflow inside | | save_workflow_json | off | Writes the workflow beside each image, numbered to match | | save_summary_txt | off | Readable summary: model, LoRAs, sampler, steps, CFG, seed, prompts | | shorten_if_needed | on | Only acts if the path would break the Windows limit | | model_name | "" | Force a name instead of detecting one |

Share the picture, keep the recipe

Set embed_workflow off and save_workflow_json on:

art_krea2_turbo/
├─ draft_krea2_turbo_00007_.png    <- clean, no metadata, safe to post anywhere
└─ draft_krea2_turbo_00007_.json   <- the workflow, kept on your disk

ComfyUI's own switch for this is --disable-metadata, which is global, applies to every save node, and would remove the workflow entirely rather than moving it beside the file. This toggle is per-node.

Drag the .json into ComfyUI to reload the graph, exactly as you would the PNG.

Why the numbers match

The .json and .txt carry the same counter as the image because this node does the saving and therefore knows which number it landed on. A node that only hands SaveImage a filename prefix has no way to know that, which is why Model Name Extractor cannot offer this.

Path shortening

shorten_if_needed measures the finished path against Windows' 260-character limit and does nothing unless it would be exceeded. When it must act it first drops suffixes that describe the file rather than the model - _fp8_scaled, _fp16, _pruned_emaonly and similar - and only truncates if that is still not enough. If even truncation cannot make it fit, it says so in the log instead of letting the save fail silently part-way through a batch.

This matters more than it sounds: create_folder puts the model name in the path twice, so every character saved counts double.

---

Model Name Extractor

Names your saved files after the model that made them. Switch checkpoints and the filename follows, with nothing to retype.

Load Checkpoint ──MODEL──> Model Name Extractor ──STRING──> SaveImage.filename_prefix

filename_prefix is a widget by default. Right-click the SaveImage node and choose Convert widget to input before you can connect to it. This is the one step everybody gets stuck on.

Found under MacroNodes/utils.

Inputs

| Input | Default | What it does | |---|---|---| | model | - | Connect any MODEL output | | prefix | "" | Prefixes the FILE name, e.g. Draft_ | | folder_prefix | "" | Prefixes the FOLDER name, e.g. art_. Needs create_folder | | create_folder | off | On: saves into a subfolder named after the model | | model_name | "" | Type a name here to override auto-detection |

Use this rather than Save Image (Macro) when you are saving something that is not a plain image - SaveAnimatedWEBP, the VideoHelperSuite nodes, or any third-party save node. It just produces a string, so it works with all of them.

Results

With sd_xl_base_1.0.safetensors loaded:

| Settings | Output file | |---|---| | defaults | sd_xl_base_1.0_00001.png | | prefix = Draft_ | Draft_sd_xl_base_1.0_00001.png | | create_folder = on | sd_xl_base_1.0/sd_xl_base_1.0_00001.png | | both | sd_xl_base_1.0/Draft_sd_xl_base_1.0_00001.png |

Folders are created if absent and reused if present.

How it finds the name

It walks the workflow graph backwards from its own MODEL input until it hits a node carrying a filename widget - ckpt_name, unet_name, model_name, model_path, gguf_name or name. That covers Load Checkpoint, Load Diffusion Model, UNETLoader, the GGUF loaders and most custom loaders, without needing to know about any of them individually.

If the walk finds nothing it inspects the model object itself, and failing that returns unknown_model rather than raising.

Names are sanitised for both Windows and Linux: < > : " / \ | ? * and control characters become underscores, trailing dots and spaces are stripped, and Windows reserved device names (CON, PRN, COM1…) get an underscore prefix. The Windows rules are applied on every platform so filenames stay portable.

IS_CHANGED returns NaN, so the node re-evaluates every run and a checkpoint swap shows up immediately.

---

Prompt Prefix (Macro)

Keeps system prompts as .txt files on disk and offers them as a dropdown, so using a different prefix per model is one click instead of a paste.

Feed prefix_text into an LLM node's system input - comfyui-ollama's Ollama Chat, a captioner, anything that takes a system prompt as a string.

| Input | Does | |---|---| | prefix | Dropdown of the .txt files found. | | custom_file | An explicit path, overriding the dropdown. | | append | Extra text added after the file's contents, for a one-off tweak. |

Outputs prefix_text and source_path, the latter so you can record which file was actually used.

Where it looks

In order: $MACRONODES_PREFIX_DIR, then <models>/prefixes, then <ComfyUI>/user/prefixes. One .txt per prefix; the filename becomes the dropdown entry.

Editing a file takes effect on the next run. Adding or removing one needs a browser refresh, because ComfyUI builds a node's dropdown contents when it scans nodes, not when it runs them.

At startup the node prints where it looked and what it found. An empty dropdown otherwise looks identical whether the environment variable is unset, the path is wrong, or the folder is simply empty.

IS_CHANGED is the file's mtime and size, so editing a prefix invalidates the cache but an unchanged file does not force everything downstream to re-run.

---

Concat Text (Macro)

Joins up to ten inputs into one string. Two things set it apart from the concat nodes already out there.

The inputs take any type. ComfyUI will not let you drag an OLLAMA_META or OLLAMA_HISTORY output into a STRING socket, so the usual workaround is to insert a Preview as Text node purely to launder the type. These slots accept anything and serialise non-strings the same way core's PreviewAny does, so those outputs connect directly.

The delimiter can hold newlines. Escape sequences are expanded wherever they appear, so \n\n gives a blank line between blocks and \n----\n gives a rule. Some existing nodes special-case the exact string \n and pass anything else through literally.

| Input | Does | |---|---| | text1–text10 | All optional. Gaps are fine. | | delimiter | Inserted between blocks. Escapes expanded. | | labels | Prefixes each block with === text2 ===. | | trim | Strips leading and trailing whitespace per block. | | skip_empty | Drops empty blocks instead of leaving stray delimiters. |

Outputs concat and block_count. labels is worth turning on while building a chain - when the result is wrong, it tells you which input caused it.

---

Adding to this pack

Nodes live one per module in nodes/. To add one:

1. Create nodes/your_node.py defining NODE_CLASS_MAPPINGS and NODE_DISPLAY_NAME_MAPPINGS.
2. Import it in nodes/__init__.py and append it to MODULES.

The root __init__.py merges everything and warns on duplicate node ids rather than letting one quietly overwrite another. Imports are explicit rather than discovered at runtime, so the full node list is readable from one file.

What this package does not do

No network calls. No eval, exec, subprocess, os.system, pickle or __import__. No dynamic imports - nodes/__init__.py lists every module by name, so the full node list is readable from one file.

Nothing to pip install. The imports are os, json, numpy, PIL, and ComfyUI's own folder_paths, comfy.cli_args and comfy.comfy_types. numpy and PIL are already hard requirements of ComfyUI itself, so the pack adds no dependency you do not already have.

Licence

MIT - see LICENSE.