ComfyUI-988
A node pack for ComfyUI. Includes LM Studio integration for local LLM/VLM text generation (streaming, vision, custom endpoints), DaC (Divide and Conquer) algorithm v2 for optimal tile-based image upscaling with intelligent overlap and blending, and tile divide/select/combine utilities.
Nodes (17)
The node that stitches your tiled upscale back together with no visible seams
The version that fixes your tiles when a model resized them
The planner that makes giant upscales fit in your VRAM
DaC v2 adds a TileCount mode — pick how many tiles, not just how much to scale
Rescale a whole tiling plan without re-running the algorithm
The generic readout that shows any wire's value on the graph
The cutter in the DaC upscale pipeline
A local LLM/VLM inside your ComfyUI graph, no API key, no cloud
The passthrough node that frees your VRAM on its way through
Dump a whole folder into the graph as a batch
Shove a dozen loose wires into one tidy pipe
Unpack the pipe back into real sockets (and chain it onward)
Read an image's aspect ratio so you don't guess it
Turn an aspect ratio and a megapixel count into clean dimensions
A batch of reproducible seeds from one base number
A tiny DSL for making number ladders
One place to hold your sampler, scheduler and steps
ComfyUI-988
Collection of useful ComfyUI nodes.
🦇 988 Kajan nodes
Responsibilities
This project contains code adapted from:
| Original project | Author | Licence | Adapted files |
|-------------------|-------|----------|-------------------|
| EA_LMStudio | EnragedAntelope | MIT | node_lm_studio.py, node_llm_unload.py, _model_fetcher.py, _config_manager.py |
| ComfyUI_Steudio | Steudio | GPL v3 | node_dac_algorithm_v2.py, node_divide_image_select.py, node_combine_tiles_v2.py, node_dac_data_scale.py |
| ComfyUI-Photopea | coolzilj | MIT | js/photopea/ |
| Node | Subcategory | Description |
|------|-------------|-------------|
| LM Studio 988 | 🦇988/LM Studio | Text generation using local LLM/VLM models via LM Studio server. Supports vision (multi-image), reasoning extraction (DeepSeek, Qwen, QwQ, GLM, GPT-OSS), draft models for speculative decoding, and system message templates. |
| LM Unload 988 | 🦇988/LM Studio | OPTIONAL Passthrough node that unloads LLM models from LM Studio and/or clears ComfyUI VRAM. |
| DaC Algorithm 988 v2 | 🦇988/Image | Calculate optimal upscale dimensions for a minimum scale factor while maintaining tile overlap constraints. |
| Divide Image Select 988 | 🦇988/Image | Divide the upscaled image into tiles using coordinates from DaC Algorithm. tile=0 returns all tiles, tile=N returns a single tile. |
| Combine Tiles 988 v2 | 🦇988/Image | Combine processed tiles back into a single image using Gaussian-blurred overlap masks for seamless blending. |
| DAC Data Scale 988 | 🦇988/Image | Scale all DAC_DATA dimensions (width, height, tile, overlap) by a multiplier. Grid and tile order are preserved unchanged. |
Installation
cd ComfyUI/custom_nodes
git clone https://github.com/your-username/ComfyUI-988.git
cd ComfyUI-988
pip install -r requirements.txt
Structure
ComfyUI-988/
├── __init__.py # Aggregates all node mappings
├── server_routes.py # Server API routes
├── nodes/ # Python node implementations
│ ├── node_lm_studio.py # LM Studio 988
│ ├── node_llm_unload.py # LM Unload 988
│ ├── node_dac_algorithm_v2.py # DaC Algorithm 988 v2
│ ├── node_divide_image_select.py # Divide Image Select 988
│ ├── node_combine_tiles_v2.py # Combine Tiles 988 v2
│ ├── node_dac_data_scale.py # DAC Data Scale 988
│ ├── _config_manager.py # Configuration management
│ ├── _model_fetcher.py # LM Studio model cache
│ └── _dac_shared.py # DaC shared helpers
├── js/ # Frontend extensions
│ ├── lm_studio/
│ │ ├── index.js # Extension entry point
│ │ ├── model_refresh.mjs
│ │ └── templates.mjs
│ └── photopea/
│ └── index.js # Photopea integration
├── config/ # Configuration files
│ ├── default_config.json
│ ├── system_message_templates.json
│ └── user_config.json # Gitignored — local overrides
├── workflows/ # Example workflows
│ ├── DaC_Upscale.json
│ └── DaC_Upscale.jpg
├── requirements.txt
├── pyproject.toml
└── .gitignore