ComfyUI-KikoTools Introduction
ComfyUI-KikoTools is a collection of custom nodes and interface tools for ComfyUI. Its nodes appear under the ComfyAssets category. They help you choose image dimensions, manage seeds and sampling settings, compare models and LoRAs, inspect workflow data, and save finished images.
Think of KikoTools as a set of small tools for recurring creative tasks. You can use a resolution picker instead of looking up a model’s preferred sizes, keep a record of seeds you may want to revisit, or view a contact sheet while comparing checkpoints. You only need to add the tools that serve your workflow.
The author created the collection after finding that larger custom-node packages often required maintaining many features that were not needed for a particular workflow. KikoTools brings the author’s selected tools together in one maintained collection. Its shared controls support both ComfyUI Classic and Nodes 2.0, and follow ComfyUI’s light and dark themes.
How ComfyUI-KikoTools Works
A ComfyUI workflow passes information between connected nodes. KikoTools adds nodes that prepare information for other nodes, work on an image, or show you what happened during a run.
For example:
- A dimension node chooses a width and height that you can connect to a latent or image-processing node.
- A sampling helper gathers settings such as sampler, scheduler, steps, and CFG in one place.
- A display node shows text or other values so you can check what is moving through your workflow.
- An output node saves images or builds a comparison sheet from multiple runs.
Some KikoTools features also change how you interact with the canvas. Embedding Autocomplete suggests terms as you type a prompt; Custom Colors helps distinguish nodes and groups; and Follow Execution helps you find the node currently running.
A latent is the working image representation used during generation, before the final image is decoded. An empty latent sets the starting canvas size. A seed is a number used in generation; keeping the seed and other settings can help you return to a result you liked. A checkpoint is a model file, while a LoRA is an additional model asset commonly used to influence a result. KikoTools helps organize and test these assets, but it does not replace the models themselves.
ComfyUI-KikoTools Features
Choose and adjust image sizes
- Resolution Calculator: Takes an image or latent and calculates larger target dimensions. Set a scale factor from 1.0× to 8.0×; the output preserves the original proportions and uses dimensions divisible by eight. For example, scaling an 832 × 1216 portrait by 1.5× gives 1248 × 1824. Use it when planning an upscale without calculating each dimension yourself. Resolution Calculator workflow
- Width Height Selector: Choose from 26 presets covering SDXL, FLUX, and ultra-wide formats, or enter custom dimensions. Its swap button changes the orientation, such as turning a 1920 × 1080 landscape into a 1080 × 1920 portrait.
- Empty Latent Batch: Creates one to 64 empty latents at your chosen size. It shares the resolution presets and orientation swap control, and shows memory estimates and warnings for large batches. A batch of four gives your sampling workflow four starting canvases rather than one. Empty Latent Batch workflow
- Model Resolution Picker: Choose a model family, an aspect-ratio chip, and a size. The node outputs an empty image latent plus separate width and height values. Its
resolutiondropdown stays in sync with the visual picker; when you select a listed resolution, that size takes priority over manually entered width and height during the run. You can also enter a custom ratio such as 16:9, type dimensions, or swap orientation. See the Model Resolution Picker guide. Model Resolution Picker in dark mode - Image to Multiple Of: Changes an image’s dimensions to a chosen multiple, for example to make an awkwardly sized image compatible with a later step. Choose center crop when removing a small amount around the edges is acceptable, or rescale when you need to keep the full composition. An image at 513 × 769, adjusted to a multiple of 64 by cropping, becomes 512 × 768. Image to Multiple Of workflow
- Image Scale Down By: Reduces width and height together using a factor from 0.01× to 1.0×. Use it to make smaller previews or reduce the size of images passed to later workflow stages.
Manage generation settings
- Seed History: Records seed changes with timestamps and lets you click an earlier seed to restore it. It keeps up to 10 recent entries in browser storage, so you can revisit promising seeds across browser sessions. A random-seed button helps you explore; a clear button removes the stored history.
- Sampler Combo: Groups the sampler, scheduler, step count, and CFG setting in one node. Its tooltips and validation help you configure these related settings without spreading them across several controls. For instance, you might start with 20 steps and adjust them while comparing speed and results.
- Flux Sampler Params: Prepares FLUX-oriented settings such as guidance, shift, and steps. It can generate multiple parameter sets for comparison, including ranges written as
start...end+step. - Sampler Select Helper and Scheduler Select Helper: Help you explore sampler and scheduler choices, including model-aware recommendations. Treat recommendations as starting points: compare the resulting images to decide what suits your work.
- Text Encode Sampler Params: Brings prompt encoding and sampling parameters into one node. It can help you keep settings consistent while working through multiple prompts.
Compare models, LoRAs, and settings
- LoRA Folder Batch: Finds
.safetensorsLoRAs in a folder and lets you test multiple files or strength values. Include and exclude filters narrow the selection; epoch-aware sorting helps when comparing files saved during training. For example, a strength range of0.9...1.2+0.1can support a side-by-side test. LoRA Folder Batch example - Checkpoint Folder Iterator: Selects one checkpoint from a folder for each queued run and outputs its model, CLIP, VAE, name, label, index, and total count. Set its index to 0 with increment-after-generate control, then queue the number of runs you need. It loads one checkpoint at a time. If you queue beyond the final checkpoint, that run reports an error instead of silently starting the sweep again.
- Model Compare Sheet: Pairs with Checkpoint Folder Iterator. It saves each run’s images, labels the tiles, and rebuilds a contact sheet as runs finish. Adjust the number of columns, tile width, and label font size to make comparisons easier to read. The node returns a preview no wider than 1024 px, while retaining the full-resolution
grid.png. - Plot Parameters: Provides line, bar, scatter, and heatmap views for comparing settings across experiments. Use it when a visual summary makes a parameter sweep easier to interpret.
The xyz-helpers tools were adapted from comfyui-essentials-nodes.
View, load, and organize your work
- Kiko Save Image: Saves PNG, JPEG, or WebP images and can open a floating viewer immediately afterward. Choose JPEG or WebP quality from 1–100, PNG compression from 0–9, or lossless WebP when appropriate. The viewer can be moved and resized; you can open, download, or select multiple saved images. Turn its popup off for a save operation when you do not need it. Kiko Save Image workflow
- Display Text: Shows long text with wrapping, scrolling, and a copy control. When it recognizes positive and negative prompt sections, it presents them separately, making generated prompts easier to inspect and reuse. Display Text workflow
- Display Any: Shows the value of an input or, in shape mode, its dimensions. If a workflow produces an unexpected result, connect this node at a point you want to inspect and check what is actually arriving there. Display Any workflow
- Local Image Loader: Offers a thumbnail gallery for browsing local images, video, and audio files. You can navigate folders, sort and page through large collections, preview media, and return the displayed folder path for another workflow step. It can also read embedded prompt and workflow information from generated images.
- Model Downloader: Downloads assets from Civitai, Hugging Face, or a direct URL within a workflow. It reports progress and supports cancellation. Some assets require an access token: the readme recommends keeping Civitai and Hugging Face tokens in the ComfyUI server environment, rather than typing them into the node, because a token entered in the node is saved with the workflow. See the Civitai API key settings and Hugging Face access-token settings for account-specific access.
- Project File Path: Builds an organized prefix for a save node, such as
MyProject/images/image. You can add a date folder, specify nested subfolders, and choose a path separator. It removes unsafe path elements so the resulting prefix stays within the output folder. - Workflow Timer: Displays elapsed workflow time while a run is in progress and keeps the final time visible afterward. Its color and optional glow are controlled in KikoTools settings. Place timers at different points if you want to compare how long stages take.
- Kiko Purge VRAM: Helps manage GPU memory between demanding workflow stages. Choose
soft,aggressive,models_only, orcache_only, and optionally set a memory threshold so purging happens only above that value. Its input passes through unchanged, while its report can show memory use before and after the operation. It also handles environments without CUDA. - Film Grain: Adds film-like texture to an image. Adjust grain size and intensity for a subtle or pronounced effect; color saturation, shadow response, and a red-channel multiplier provide further control. It preserves transparency when present.
Customize the ComfyUI interface
- Embedding Autocomplete: Suggests embeddings after
embedding:, LoRAs after<lora:, and custom tags as you type. You can change triggers, choose completion keys, add a comma automatically, replace underscores with spaces, and load custom word lists. The tag-pack guide covers available packs. This feature builds on the autocomplete work in ComfyUI-Custom-Scripts.
<img src="https://github.com/ComfyAssets/ComfyUI-KikoTools/blob/main/examples/ac-emb.png?raw=true" width="30%" alt="Embedding suggestions" /><img src="https://github.com/ComfyAssets/ComfyUI-KikoTools/blob/main/examples/ac-lora.png?raw=true" width="30%" alt="LoRA suggestions" /><img src="https://github.com/ComfyAssets/ComfyUI-KikoTools/blob/main/examples/ac-tag.png?raw=true" width="30%" alt="Tag suggestions" /> - Custom Colors: Right-click a node and choose Custom Colors to change its full color, title, or background. You can apply a change to several selected nodes or color a group. Colors are saved with the workflow. See the dark-theme and light-theme examples.
- Follow Execution: Enable Follow Execution: Enable in ComfyUI settings, then use the canvas context menu to start or stop following the running node. You can also show the executing node or navigate by node type. Auto-start begins following when a workflow starts. See the Classic and Nodes 2.0 menu screenshots.
Gemini Prompt Engineer is deprecated. The node is hidden from normal new-node discovery but remains available so existing workflows using its GeminiPrompt identifier can load. It still depends on Google’s service and the optional google-genai package; future API compatibility is not guaranteed. See the legacy Gemini documentation for its existing behavior.
ComfyUI-KikoTools Models
KikoTools does not provide its own image-generation models. Its model-related tools help you choose dimensions or compare model files you already use.
The Model Resolution Picker ships size tables for 14 model families. Families explicitly named in the provided documentation include SD 1.5, SDXL, Illustrious XL, Flux.1, Flux.2, and Qwen-Image. Choose the family matching your workflow to see its listed aspect ratios and sizes. For example, changing from one family to another changes the sizes offered by the picker; it does not load or switch the generation model.
The picker also includes first/last-frame size rows for Wan 2.2, LTX-2, HunyuanVideo 1.5, and MiniMax H3. These rows help select dimensions for video workflows, but the picker’s latent output is an image latent and must not be fed to a video model as though it were a video latent.
For a complete explanation of the shipped families and the optional file for adding, replacing, or hiding size entries, consult the Model Resolution Picker documentation. To compare actual checkpoint files, use Checkpoint Folder Iterator with Model Compare Sheet; to compare LoRAs, use LoRA Folder Batch.
What’s New with ComfyUI-KikoTools
- Version 1.3.0 — Model Resolution Picker: Added a visual aspect-ratio picker, a synchronized native
resolutiondropdown, shipped family-size tables, and an optional user override file. You can choose dimensions without memorizing size lists, while still entering your own. See the release notes and node guide. - Version 1.2.0 — unified ComfyAssets UI: Brought KikoTools controls into a consistent interface that follows ComfyUI’s light and dark themes and works in Classic and Nodes 2.0. Existing workflow interfaces and settings were retained. See the release notes, UI migration overview, and screenshots.
The supplied readme does not describe earlier versions individually; the changelog is the appropriate place to check their release details.
Troubleshooting ComfyUI-KikoTools
A dimension or resolution is not what I expected.
- Check which preset, aspect-ratio chip, or model family is selected.
- In Model Resolution Picker, check the
resolutiondropdown. A listed resolution takes priority over typed width and height when the node runs. - If you intend to use your own dimensions, select
custom, enter the width and height, and run the workflow again. - Remember that some dimension tools adjust sizes to the required step or multiple. Check their output width and height rather than assuming the typed values were used unchanged.
The Model Resolution Picker ignores my custom family sizes.
- Review the optional override file at
ComfyUI/user/kikotools/model_resolutions.jsonunder the default user directory. - Check the picker panel for an explanation: an invalid file is ignored.
- Compare your file with the override-file guidance, correct the problem it reports, and try again.
A checkpoint sweep stops with an index error.
- Check how many checkpoints the iterator found.
- Start at index 0 and set
control_after_generateto increment. - Queue a run count equal to the checkpoint count. Runs beyond the last checkpoint intentionally fail rather than wrapping around.
The comparison sheet contains images from an earlier attempt.
- Start a new Checkpoint Folder Iterator sweep at index 0.
- Use the same
run_namefor the tiles belonging to that sweep. - Check the new sheet after runs finish. At index 0, Model Compare Sheet archives tiles already in that run folder so they are not included in the new grid.
A model download requires authentication.
- Check whether the Civitai or Hugging Face asset requires account access.
- Obtain the appropriate token from that site.
- Keep the node’s
api_tokenempty if you can set the token where the ComfyUI server starts: useCIVITAI_API_KEYfor Civitai orHF_TOKENfor Hugging Face, then restart ComfyUI. - Avoid sharing a workflow containing a token typed into the node, because that value is saved with the workflow.
I cannot find Gemini Prompt Engineer when adding a new node.
The node is deprecated and intentionally hidden from normal new-node discovery. It remains for previously saved workflows. Consult the legacy documentation if you are maintaining one of those workflows.
Frequently asked questions
- Does KikoTools include SDXL, FLUX, or other checkpoints? No. Its size picker includes information for model families; model files themselves are separate assets.
- Can I use it in ComfyUI Classic and Nodes 2.0? Yes. The shared ComfyAssets controls support both interfaces.
- Does Seed History keep every seed forever? No. It keeps up to 10 recent entries in browser storage.
- Can I use the video-family rows to create a video latent? No. Those rows provide first/last-frame dimensions; the picker outputs an image latent.
- Will Kiko Purge VRAM change the data passing through my workflow? Its input is passed through unchanged. Its purpose is memory management, with an optional usage report.
Learn More about ComfyUI-KikoTools
Start with the guide for the tool you are using:
- Model Resolution Picker, Resolution Calculator, Width Height Selector, and Empty Latent Batch for planning image sizes.
- Seed History, Sampler Combo, and Flux Sampler Params for generation settings.
- LoRA Folder Batch and the LoRA testing example workflow for comparing assets.
- Kiko Save Image, Display Text, Display Any, and Local Image Loader for reviewing results and workflow data.
- UI screenshots for examples of controls in ComfyUI, and the changelog for release updates.
The linked example images throughout this readme show individual tools in context. Use them alongside the node guides when deciding where a tool belongs in your own workflow.
