ComfyUI-NO8D-controls

English | 简体中文

This custom node pack provides an end-to-end optimization solution covering the entire workflow from image loading to image generation. It redesigns key nodes for image generation, LoRA stacking, prompt generation, and more to simplify ComfyUI workflow construction, improve generation efficiency across nodes, and significantly lower the barrier to building and using workflows.

Quick navigation

Installation · Nodes · Prompt API & LLM · More tips · License · Star History

Installation

Clone the repository into ComfyUI/custom_nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/no8d/ComfyUI-NO8D-controls.git

ComfyUI Manager installs the dependencies from requirements.txt automatically. If you clone or update the node manually, install them with ComfyUI's Python environment:

cd ComfyUI/custom_nodes/ComfyUI-NO8D-controls
python -m pip install -r requirements.txt

For Windows Portable, run this from the portable root instead:

python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-NO8D-controls\requirements.txt

openpyxl is required for importing and exporting XLSX prompt libraries. Do not install it into an unrelated system Python environment.

Restart ComfyUI and hard-refresh the browser page. No frontend build step is required.

An example workflow is included at examples/NO8D-controls-example.json.

The Prompt Libraries example is available at examples/NO8D-Prompt-libraries.json.

You can also follow NO8D on Patreon for project updates.

Nodes

All nodes are available under the NO8D-control or NO8D-controls category.

NO8D-Prompt-libraries

Browse, manage, and combine reusable prompt libraries visually.

• Starts empty; eight optional XLSX libraries are bundled in data/krea_style_libraries/ for manual import.
• Imports TXT, CSV, XLSX, and public Google Sheets, with search, favorites, history, editing, and portable export.
• Supports single, multiple, all, or random card output and preserves prompts exactly as stored.
• Supports mouse navigation and Arrow, PageUp, PageDown, Home, and End shortcuts.

📖 Read the NO8D-Prompt-libraries tutorial on Patreon.

NO8D-Prompt Merge

Merge outputs from multiple Prompt Libraries or other text nodes into one prompt, following the visible input-port order.

NO8D-LoRA stack

Load a UNET through ComfyUI's native loader, then manage and apply multiple LoRAs without a CLIP input.

• Add, remove, enable, disable, and reorder LoRAs.
• New LoRAs start at strength 1.0 with a default 0–2 slider range.
• Merge trigger words from enabled LoRAs into one text output.
• The UNET controls are inherited from ComfyUI's native loader and follow upstream updates.

NO8D-Prompt

Expand text, analyze a reference image, or combine both into a complete positive prompt through a configured API.

Before first use: This node is an API client and does not include an LLM. Configure an OpenAI-compatible API or a local Ollama service first. See Prompt API and LLM configuration.

• Supports text-only, image-only, and text-plus-image input.
• Provides style, shot-scale, and prompt-length controls, including a no-style-description mode.
• Keeps the source medium separate from a locked target style and corrects conflicting style vocabulary once when needed.
• Allows separate text and vision model selection.
• Text-only prompting requires a text-capable LLM; reference-image analysis requires a vision-capable multimodal model.

NO8D-Prompt-view

Display and edit prompt text before sending it downstream.

• Automatically displays incoming prompt text.
• Supports a fixed prompt that is prepended whenever the node emits text.
• Supports manual editing and one-click downstream sending.
• Can temporarily stop automatic text output without losing the editor content.

NO8D-Load-images

Load and organize one or more local images as a ComfyUI list output.

• Add images by file picker, drag-and-drop, or clipboard paste.
• Automatically selects the first image in each newly imported batch.
• Select, reorder, and preview individual images.
• Sends only selected images downstream; selecting all outputs all images, while selecting none produces no output.

NO8D-Generate

Combine ComfyUI sampling controls, image preview, inpainting, and outpainting in one compact node.

• Controls sampler, scheduler, steps, CFG, denoise, and seed.
• Supports editable canvas ratios, portrait/landscape swapping, source-image positioning, lasso masks, feathering, opacity, invert, and clear controls.
• Automatically switches between normal generation, native inpainting, and outpainting according to the canvas state.
• Supports native reference workflows for Flux2 Klein and Krea2 Identity Edit while preserving the connected latent for ordinary generation.
• Outputs the final generated image.

NO8D-Remove Krea2 Reference Latents

Remove Ostris Krea2 reference_latents from conditioning while preserving the prompt, visual-language conditioning, and other metadata.

NO8D-Match Krea2 Reference Latents

Resize Ostris Krea2 reference-latent grids to match a target latent while preserving the remaining conditioning metadata.

NO8D-A/B preview

Compare two image streams with an interactive split preview.

• Displays image A and image B with their original dimensions.
• Supports list-page switching and single-stream history comparison.
• Provides separate image_a and image_b outputs; the auto-output switch controls the image A branch.
• Native ComfyUI image actions target the A or B image selected under the right-click pointer.

NO8D Image Grid

Combine multiple image inputs or image batches into one image.

• Arrange images horizontally, vertically, or in an automatic grid using the first image as the size reference.
• Standard, left, center, and right crop modes keep the combined image content at the first image dimensions.
• Crop modes expand the canvas for outer margins and center spacing, preserving image content and filling added areas with the background color.
• Cropped images follow the layout direction: horizontal splits left/right, while vertical and one-column grids split top/bottom.
• Use one value for the outer margin and image spacing; grid padding uses a darkened background color.

NO8D Image Title

Add title bars outside the top or bottom edge, or overlay a title across the middle.

• Set independent batch titles with one line per image; blank titles skip title rendering.
• Resize by the longest edge to the original size, 1K, 2K, or 4K before rendering the title, while preserving aspect ratio.
• Control the title background, opacity, position, height, font size, padding, text color, and alignment.
• Top and bottom bars increase output height; middle titles preserve the source dimensions.
• Inner top and inner bottom positions overlay the image edges without changing output dimensions.
• Title-bar opacity and height use percentages; bar height is relative to the source image height.

NO8D save

Save images and matching captions for image-text datasets.

• Builds filenames from fixed text, source filename, date/time, and size class.
• Supports drag-to-reorder filename parts.
• Saves caption text alongside each image.

NO8D-Empty latent

Choose dimensions and create the latent through ComfyUI's native EmptyLatentImage node.

• Provides portrait and landscape aspect ratios with short-side presets and manual short/long-side overrides.
• Existing saved values are migrated from the previous model-family and width/height controls.
• Outputs the latent together with its calculated width and height.

Prompt API and LLM configuration

NO8D-Prompt sends your text or reference image to a configured large language model (LLM), then returns the generated prompt. It can connect to an OpenAI-compatible API or a locally running Ollama service.

1. Open the API Manager

Open ComfyUI Settings, select NO8D-control in the left sidebar, then click API Manager under Prompt API configuration in the Prompt panel.

2. Add and validate a service

1. Select an existing service on the left, or click + Add service.
2. Choose OpenAI-compatible API or Local LLM (Ollama) as the service type.
3. Enter the provider's Base URL and API key if required. A local Ollama service does not require an API key.
4. Click Validate API. After validation succeeds, choose the available text and vision models.
5. Click Save.

3. Choose the right model

• Text model: used for text expansion and text-only prompt generation.
• Vision model: used for image-only and text-plus-image prompting; it must support image input.
• A multimodal model may be selected for both roles if it supports text and image input.

After saving, return to the Prompt settings page and select the service under Default prompt API. If no models appear, check the Base URL and API key, then validate the service again. For Ollama, make sure Ollama is running and the required model is already installed locally.

API keys remain in the local ComfyUI environment and should never be committed to the repository.

More tips and experimental nodes

For more usage tips, workflow ideas, and experimental NO8D nodes, visit Patreon. You can also follow ongoing updates and support the project's continued development there.

License

MIT. See LICENSE.

Star History