Uninodes

Uninodes

Uninodes is a collection of custom ComfyUI nodes for building prompts, generating images, finishing them, and comparing the results — without needing a large workflow.

Current version: 1.0-ish

Nodes

| Node | What it does | | ---------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Prompt Composer | Build positive and negative prompts from reusable blocks and preset libraries. Chain composers together; optionally output encoded conditioning when a CLIP is connected. | | Direct Sampler | Generate images in one node. Choose a checkpoint or diffusion model, configure sampling and resolution, add LoRAs, and use text-to-image or image-to-image. Includes image preview and saving controls. | | Image Finisher | Process an existing image with resizing, model upscaling, tiled diffusion refinement, face/eye/hand detailing, mirroring, and text/image watermarks. Compare the original and processed images and save either or both. | | Direct Image Sampler Finisher Ultra Turbo Deluxe | The all-in-one version: Sampler settings → Finisher settings → Compare & Save, arranged in three columns. Generates an image, optionally finishes it, and outputs both the final and original images. Yes, that's its full name. |

All four nodes are available under the Uninodes category in ComfyUI.

Tool

You'll find in tools/ a simple html page you can open. It allows you to open/edit or even create your own libraries. Libraries accepts variables. Variables can be free input or connected libraries (for exemple, for colors).

Installation

1. Install and run a recent version of ComfyUI.

2. Download this repository as a ZIP from GitHub (Code → Download ZIP) or clone it.

3. Copy the Uninodes folder into ComfyUI/custom_nodes/ so the resulting path is:

   ComfyUI/
   └── custom_nodes/
       └── Uninodes/
           ├── __init__.py
           ├── backend.py
           ├── web/
           └── ...
   

4. Restart ComfyUI and refresh the browser (Ctrl+F5 if the controls don't appear).

5. Right-click on the canvas and find the nodes under Uninodes.

Dependencies

Basic features

• ComfyUI (with its usual Python dependencies, including PyTorch, NumPy, Pillow and aiohttp).
• Image model weights (checkpoints or diffusion models). These are not included in the repository.

There is no separate pip install step for the basic nodes when ComfyUI's normal environment is already set up.

Optional features

| Feature | Additional requirement | | ---------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Face, eye and hand detailing | ComfyUI Impact Pack, a compatible detector provider (often Impact Subpack), and detector model weights. A SAM model can be used if configured. | | Tiled diffusion refinement | ComfyUI Ultimate SD Upscale (the UltimateSDUpscaleNoUpscale node). | | AI model upscaling | A compatible upscaler model in ComfyUI/models/upscale_models/. |

You only need these extras when enabling the corresponding feature. Simple generation, resizing, comparison, and saving do not require Impact Pack or Ultimate SD Upscale.

Models

Direct Sampler can list models from:

• ComfyUI/models/checkpoints/
• ComfyUI/models/diffusion_models/

It supports conventional SD/SDXL-family checkpoints (including Pony and Illustrious), and provides loading paths for several model-only architectures such as Anima, FLUX, SD3/3.5, Qwen Image, Z-Image, HiDream, and others supported by your ComfyUI installation.

Some architectures need separate text encoders and VAEs in models/text_encoders/ and models/vae/. Use automatic companion loading where available, manually select the matching encoders, or connect external MODEL / CLIP / VAE inputs. Uninodes does not download model weights.

Note: Model loading is not a guarantee that one sampling preset works for every architecture. Some models require different guidance, schedules, or specialized workflows.

Quick start

For a simple image-generation workflow, add Direct Sampler, select a model, enter a prompt, adjust your settings, and queue the prompt.

To process an existing image, connect it to Image Finisher and enable only the operations you want.

For the combined workflow, use Direct Image Sampler Finisher Ultra Turbo Deluxe: configure generation in column 1, finishing in column 2, and preview/compare/save in column 3. Its image output is the processed result (or the original when finishing is disabled); original always preserves the generated image.

Notes

• Extra detailers and specialized model families depend on the versions of the third-party nodes and ComfyUI installed on your system.
• GPU execution and Node 1.0/2.0 rendering can vary by setup; please include your ComfyUI version, model family, and full error log when reporting issues.