AUN-ComfyUI-Nodes Introduction
<img src="docs/images/aun-logo.png" alt="AUN logo" height="100" />
AUN-ComfyUI-Nodes is a collection of custom nodes for ComfyUI. It helps you build image and video workflows that are easier to control, inspect, and keep organized.
If your canvas has become crowded with repeated settings, long connection lines, or separate workflows for every prompt variation, these nodes offer ways to bring those tasks together. You can switch prompts and LoRAs by index, manage groups of nodes, prepare generation settings in one place, compare results, and save files with descriptive names.
Many AUN nodes have a compact mode that hides configuration widgets while keeping the node useful on the canvas. Selected nodes also support Collapse Connections, which visually gathers their connection lines without disconnecting them.
How AUN-ComfyUI-Nodes Works
ComfyUI workflows pass information between connected nodes. AUN nodes work within that system: some prepare information, some choose which information to pass along, and others display or save the result.
A useful way to think about an AUN workflow is as a small production line:
- Choose the ingredients. An Inputs node can provide a model, an empty latent image, and sampling settings. A prompt switch can choose the text for this run.
- Coordinate the choices. The prompt switch can output an index, which is simply a number identifying the selected prompt. A LoRA node can use that same number to choose the LoRAs intended for that prompt.
- Generate and inspect. A sampler creates the result. Preview nodes can show the active prompt, filename, images, or other values without requiring you to trace every wire.
- Save the result. Filename and save nodes can include details such as the model, seed, steps, or CFG in the output name.
For example, you could keep several portrait prompts in one workflow. Each time the prompt changes, its index tells a LoRA loader which style additions to apply. After generation, an image comparer lets you inspect the new result beside an earlier version.
Compact mode changes how a node looks, not what it does. Similarly, Collapse Connections changes how sockets and lines appear on the canvas; the underlying connections remain in place.
AUN-ComfyUI-Nodes Features
Keep complex canvases manageable
- AUN Node Controller manages bypass, mute, and collapse states for up to 20 slots. You can target nodes by ID or title, use an all-at-once switch, or select a slot with an index from another node.
- AUN Group Controller provides similar controls for ComfyUI groups. It can display all discovered groups or a selected list.
- Title, group, and index controllers provide more focused ways to bypass or mute parts of a workflow. These are useful when you want to compare alternate processing paths without rebuilding their connections.
- Collapse Connections Controller can visually simplify selected nodes. Its graph-wide action can collapse connections on nodes with multiple slots. This controller requires the experimental Global collapse connections setting under Settings → AUN.
- AUN Bookmark lets you assign a key press to a position in a large workflow so you can return to it quickly.
Switch labels help make states readable at a glance: for example, Active 🟢, Bypass 🔴, and Mute 🔇 indicate whether a controlled path is participating in the workflow.
Select and build prompts
- AUN PromptCycler moves through prompts sequentially, randomly, manually, by a specified range, or by search. You can give a prompt a recognizable title using
Title: Prompt text. - AUN Multi Prompt Cycler returns all prompts matching a range or search, allowing each match to trigger downstream execution. Its range syntax includes entries such as
1,2,4-8,11;0selects all prompts. - Text Index Switch 4 chooses among up to 20 text inputs using Select, Increment, Random, or Range mode. It also outputs the chosen label and index, which can control other nodes.
- Text Index Switch 5 additionally reads supported
key=valuesettings from the selected text, such asseed,steps, andcfg. It sends those values to separate outputs and removes the tokens from the prompt text. A Diffusers variant supports diffusion, CLIP, and VAE selections. - Add-To-Prompt and Add-To-Prompt Multi place optional text before or after your main prompt. The multi version supports up to ten individually controlled additions. For instance, you can keep a lighting phrase available without adding it to every image.
- AUN Wildcard Add-To-Prompt varies text written with wildcard syntax such as
__name__or{a|b|c}on each execution. - Negative Prompt Selector chooses from ten preset negative prompts.
- Keyword Preset Selector chooses a preset when words in a reference phrase match a configured row. The first matching row wins, so put more specific matches above broader ones.
Match LoRAs to your ideas
A LoRA is an optional addition applied to a model to influence its output. AUN offers several ways to choose and combine them:
- Random LoRA Model Loader selects one LoRA from up to ten slots. Select, Increment, Random, and Range modes let you fix a choice or vary it between runs. Slots can include trigger text and strength settings.
- LoRAs by Prompt Index associates up to three LoRAs with each of up to 20 prompt indices. Connect a prompt selector’s index output so the selected prompt and LoRAs stay together. Its setup view hides empty LoRA slots for a tidier display.
- Random Multi-LoRA Model Loader provides the earlier multi-prompt arrangement, including per-slot controls and trigger-word display.
- LoRA Stack With Triggers Model Clip combines up to ten LoRAs, with trigger words and separate model and CLIP strengths for each slot.
- LoRA Loader Model Only (String) accepts a LoRA name or path supplied as text, useful when another node chooses the name dynamically.
Changing a LoRA’s strength changes how strongly it influences the generation. Keeping prompt text, trigger words, and LoRA selection together can make a series of experiments easier to repeat and understand.
Prepare image generation
- AUN Inputs Basic combines checkpoint selection, sampler settings, and an empty latent image in one node. Connect its outputs to a standard KSampler pipeline when you want a compact starting point.
- AUN Inputs also prepares information used for naming and saving. The original readme describes the fuller Inputs-style nodes as a legacy direction for new workflows and recommends basic input nodes paired with AUN Save Image V2.
- Inputs Basic + Prompt Switch adds prompt selection to the basic setup. Supported
key=valuetext tokens can override matching loader settings. - Diffusers, Hybrid, and Refine Inputs variants support separate diffusion-model components, a choice between checkpoint and diffusion loading, or an optional second model for refinement.
- AUN Empty Latent creates a blank starting latent at your chosen dimensions, with aspect-ratio, width/height swap, and batch options.
- KSampler Plus provides a two-pass generation process with latent upscaling, optional pixel-space upscaling, and refinement. KSampler 2-Model can use a second model for its latent-upscale pass.
A latent is the working representation used during generation, before the image is decoded for viewing. Upscaling that latent gives the sampler another pass at a larger working size; the Image Slider Comparer can then help you judge whether the extra pass improved your result.
Load, resize, compare, and save images
- Image Loader outputs an image, mask, original filename, and cleaned filename. Load Image Single/Batch 3 can use one uploaded image or select images from a folder by increment, random choice, range, or filename search.
- Load & Resize Image and Resize Image offer resizing strategies that maintain aspect ratio and include FramePack bucket sizing and fill/crop placement.
- Img2Img helps switch between text-to-image and image-to-image preparation, including image loading, resizing, and latent encoding.
- Image Slider Comparer accepts up to five left/right image pairs. Drag the divider to inspect differences, choose a pair from the Pair dropdown, and use Frame for batched images. Its right-click menu provides actions for opening or downloading either side. With
save_activeenabled, it can save the displayed frame using a configurable prefix. - Path Filename V2 builds image paths and names. You can choose which generation details appear in a filename and generate a text or JSON sidecar describing the main parameters.
- Save Image V2 accepts a combined
path_filenameinput and supports filename customization and metadata embedding. Filename Resolver V2 helps connect AUN filename builders to standard save nodes.
For example, turning on model, seed, and steps in a filename can help you identify how an image was made later. Turning some of those details off produces shorter names when you do not need them.
Work with video and frames
- Inputs Wan2.2 Basic prepares high- and low-noise expert models and settings for Wan2.2 MoE, with options for image-to-video inputs and RIFE frame interpolation.
- Inputs MiniMaxH3 Basic prepares the components and settings used by the example MiniMax-H3 video workflows.
- AUN RIFE creates intermediate frames between existing frames. A multiplier from 2 to 10 can make motion appear smoother or produce more frames for slow-motion work. Turning
enableoff passes frames through unchanged;ensemblemay improve quality but takes longer. - Path Filename Video V2 builds video save paths for AUN save nodes. Path Filename Video (Resolved) produces a resolved filename that can be used with VHS Video Combine.
The original readme marks AUN’s older Save Video nodes as deprecated in favor of VHS Video Combine.
Inspect and route workflow data
- Show Any Multi displays up to 20 mixed inputs in one node, including text representations and inline image previews. It labels previews using connected output slots.
- Passthrough Any Multi provides a similar display and sends text representations onward through
STRINGoutputs. Use it when another node needs the displayed information as text. - AUN Any Index Switch selects one of up to 20 inputs of different types. Only the selected input is evaluated.
- Preset Manager stores keyword-triggered sets of widget values. Apply Preset To Node can apply selected values to a target node at queue time while leaving linked inputs untouched.
- Merge Bundles, Split Bundle, and Wireless Bundle help carry named values together, then expose individual fields where they are needed. These can reduce wiring in workflows that pass many related settings.
- Scan And Show Widgets displays values from a node selected by ID or title and can expose those values as outputs. Run the workflow once to populate its output slots.
AUN-ComfyUI-Nodes Models
AUN-ComfyUI-Nodes is primarily a node collection, not a single image-generation model. Its loader and sampler nodes work with models you choose for your ComfyUI workflow. The supplied context describes these model-related options:
| Option | What it is for |
|---|---|
| Checkpoint-based Inputs | Use a standard checkpoint workflow where the model, CLIP, and VAE are loaded together. |
| Diffusers Inputs | Use a standalone diffusion model with separate CLIP and VAE files. |
| Hybrid Inputs | Choose between a standard checkpoint and separately loaded diffusion components. |
| Refine Inputs / KSampler 2-Model | Use a separate model for a refinement or second sampling stage. |
| Wan2.2 high- and low-noise experts | Use the paired experts with AUN Wan2.2 MoE for the video workflows described in the original readme. |
| MiniMax-H3 components | Prepare the UNET, minimax CLIP, video VAE, audio VAE, and optional turbo LoRA used in the provided video examples. |
RIFE rife47 / rife49 weights | Create frames between existing video frames. The readme states that weights are downloaded to ComfyUI/models/rife on first use. |
These options are not interchangeable presets for the same output. Choose the loader that matches the kind of model files and workflow you are using. For example, a checkpoint-based portrait workflow calls for different preparation than a Wan2.2 video workflow.
Troubleshooting AUN-ComfyUI-Nodes
A Collapse Connections controller shows a warning
- Open Settings → AUN.
- Enable the experimental Global collapse connections setting.
- Return to the controller and check whether the warning clears.
This setting is specifically required for the Collapse Connections controller. Ordinary compact mode and ComfyUI’s built-in title-bar collapse are different features.
The wrong prompt or LoRA is selected
- Check the prompt selector’s active index output.
- Confirm that it is connected to the LoRA node’s
prompt_indexinput, if you want them synchronized. - Check the selector mode. Select keeps a fixed choice; Increment, Random, and Range can change it between runs.
- In LoRAs by Prompt Index, inspect the LoRA slots configured for the selected prompt index.
A saved image has an unexpected name
- Inspect the switches in Path Filename V2 to see which settings are included.
- Check its
path_filenameconnection to Save Image V2. - Connect the save node’s
filenameandsidecar_textoutputs to Show Any Multi to see the generated values in the workflow. - If you are using a standard save node instead, check whether Filename Resolver V2 is needed between the AUN filename builder and that node.
Show Any Multi displays nothing useful yet
- Connect the value you want to inspect to one of its inputs.
- Queue the workflow so the connected node produces a value.
- If the node is visually condensed, double-click its body to switch display mode. Use its right-click menu to show or hide data-type badges.
Scan And Show Widgets has no populated output slots
- Check that it targets the intended node by ID or title.
- Queue the workflow once. The node populates its dynamic output slots after a run.
- Use its widget selection or filter controls to narrow the displayed values if there are too many.
A previously saved Refine Inputs workflow looks different after loading
The original readme notes that changes to the SpeedLoRA-related inputs may affect existing workflows using AUN Inputs Refine or AUN Inputs Refine Basic. Check those widgets and reconnect any affected inputs before running the workflow.
Frequently asked questions
Does compact mode remove settings or connections?
No. It hides configuration widgets to make the canvas easier to scan. Collapse Connections likewise changes the visual presentation of sockets and lines while preserving connections.
Can I compare more than two images?
Image Slider Comparer supports up to five pairs. Select the pair you want to view from its Pair dropdown.
Can I use AUN nodes with a standard KSampler?
Yes. The provided examples connect AUN Inputs Basic or AUN Inputs to a standard KSampler pipeline.
Do I need a separate workflow for every prompt?
Not necessarily. PromptCycler and the text index switches can select different prompts within one workflow. An index-driven LoRA node can change LoRAs along with them.
Should I use the older image and video save nodes for a new workflow?
The original readme recommends Save Image V2 for images and marks the older AUN image saver as deprecated. It also marks AUN’s video savers as deprecated in favor of VHS Video Combine.
Learn More about AUN-ComfyUI-Nodes
The example workflows are a practical place to start: their linked preview images show the canvas layout, and the original readme says you can drop the workflow images into ComfyUI to load them.
- Prompt cycling with index-matched LoRAs shows how a selected prompt can choose its corresponding LoRAs.
- AUN Inputs workflow shows an all-in-one generation setup.
- Automatic image-saving pipeline shows filename preparation and Save Image V2 together.
- Prompt selection workflow combines positive prompts, negative prompts, and optional additions.
- Image Slider Comparer workflow demonstrates side-by-side image inspection.
- KSampler Plus workflow compares a base result with an upscaled result.
- Node controllers workflow illustrates ways to coordinate nodes and groups.
- Collapse Connections and Use Everywhere notes cover the documented interaction with cg-use-everywhere.
For video, the provided JSON examples include MiniMax-H3 reference-to-video, MiniMax-H3 text-to-video, Wan2.2 text-to-video with RIFE, and Wan2.2 image-to-video with RIFE.
