# ComfyUI Diffusers Utils ⚠️ **EXPERIMENTAL NODES - UNDER ACTIVE DEVELOPMENT** ⚠️ A set of nodes which provide flexible inference using diffusers in ComfyUI environment. **Latest Update (Jan 16, 2026):** Added support for GLM Image models (both text-to-image and image-to-image) with dedicated installation instructions. ## Current Supported Models: - LongCat 6B Image - LongCat 6B Image Edit - GLM Image (Text-to-Image) - GLM Image (Image-to-Image) ## Installation 1. Clone this repository into your ComfyUI custom_nodes directory: ```bash cd ComfyUI/custom_nodes git clone https://github.com/yourusername/Comfyui-DiffusersUtils.git ``` 2. Install the required dependencies: ```bash cd Comfyui-DiffusersUtils pip install -r requirements.txt ``` 3. For GLM Image support, also install the following packages: ```bash pip install git+https://github.com/huggingface/transformers.git pip install git+https://github.com/huggingface/diffusers.git pip install git+https://github.com/huggingface/peft.git ``` 4. Restart ComfyUI ## Available Nodes ### Pipeline Loading Node #### Diffusers Pipeline Loader (`DiffusersPipeline`) Initializes different LongCat diffusion pipelines with dynamic components. - **Inputs:** - `pipeline_class_name`: Select pipeline type ("LongCatImagePipeline" or "LongCatImageEditPipeline") - `model_path`: Path to the model directory - `pipeline`: Existing pipeline (optional, for component swapping) - `torch_dtype`: Data type (float16, float32, bfloat16) - Default: bfloat16 - `components`: Comma-separated list of components to load (optional) - `presets`: Predefined component sets to load - Default: "text_encoder, tokenizer, text_processor" - **Outputs:** - `pipeline`: The loaded pipeline object - **Category:** Diffusers/LongCat ### Text Encoding Node #### Diffusers Text Encode (`DiffusersTextEncode`) Encodes text prompts using the pipeline's text encoder. Supports both text-only and text+image encoding depending on the pipeline type. - **Inputs:** - `pipeline`: The loaded pipeline object - `prompt`: Text prompt to encode (multiline, default: "Masterpiece, best quality, 8k uhd, photo realistic,") - `image`: Optional image input for pipelines that support text+image encoding - `batch_size`: Number of images to generate (default: 1, min: 1, max: 64) - **Outputs:** - `diffusers_cond`: Conditioning information for sampling - **Category:** Diffusers/Encode ### Sampling Node #### Diffusers Sampling (`DiffusersSampling`) Performs the generation loop using a pipeline and conditioning to generate images using the diffusion process. - **Inputs:** - `pipeline`: Loaded diffusion pipeline - `diffusers_cond`: Conditioning from DiffusersTextEncode node - `steps`: Number of inference steps (default: 26, min: 1, max: 100) - `cfg`: Guidance scale (default: 1.5, min: 0.0, max: 20.0) - `negative_diffusers_cond`: Negative conditioning (optional) - `num_images_per_prompt`: Number of images to generate per prompt (default: 1, min: 1, max: 8) - `seed`: Random seed for generation (default: 42, range: 0 to 0xffffffffffffffff) - `image`: Input image for image editing (optional) - `width`: Output image width (for regular generation, or override for editing) (default: 1024, min: 256, max: 4096, step: 64) - `height`: Output image height (for regular generation, or override for editing) (default: 1024, min: 256, max: 4096, step: 64) - **Outputs:** - `image`: Generated image(s) - **Category:** Diffusers/Sampling ### LoRA Utility Nodes #### Diffusers Load LoRA Only (`DiffusersLoadLoraOnly`) Load a LoRA file without applying it to any models. Use with MergeLoraToModel or other nodes to apply the LoRA later. - **Inputs:** - `lora_path`: Path to the LoRA file - **Outputs:** - `lora`: The loaded LoRA state dictionary - **Category:** Diffusers/Lora #### Diffusers LoRA Layers Operation (`DiffusersLoraLayersOperation`) Modify specific layers in a LoRA by zeroing them out (when scale=0) or scaling them (otherwise) based on pattern matching. - **Inputs:** - `lora`: The LoRA state dictionary to modify - `layer_pattern`: Regex pattern to match layer names (default: `.*transformer_blocks\.(\d+)\.`) - `layer_indices`: Comma-separated list of layer indices to operate on, with support for ranges (e.g., '59', '10,11,12', or '50-53') (default: '59') - `scale_factor`: Scale factor to apply. Use 0 to zero out layers (default: 1.0, range: -10.0 to 10.0) - **Outputs:** - `modified_lora`: The modified LoRA state dictionary - **Category:** Diffusers/Lora #### Diffusers Save LoRA (`DiffusersSaveLora`) Save a modified LoRA state dictionary to a file. - **Inputs:** - `lora`: The modified LoRA state dictionary to save - `filename`: Filename to save the LoRA as (e.g. my_lora.safetensors) (default: "my_lora.safetensors") - `output_dir`: Directory to save the LoRA to (optional, defaults to ComfyUI output directory) - **Outputs:** - None (saves file to specified location) - **Category:** Diffusers/Lora #### Diffusers LoRA Stat Viewer (`DiffusersLoraStatViewer`) View information about LoRA layers to help define layer patterns for LoraLayersOperation. - **Inputs:** - `lora`: The loaded LoRA to analyze - **Outputs:** - `lora_info`: Information about the LoRA layers - **Category:** Diffusers/Lora #### Diffusers Merge LoRA to Pipeline (`DiffusersMergeLoraToPipeline`) Apply a pre-loaded LoRA to transformer. This allows separation of loading and applying LoRAs. - **Inputs:** - `pipeline`: The pipeline to apply the LoRA to - `lora`: The loaded LoRA to apply - `strength`: How strongly to modify the diffusion model. This value can be negative (default: 1.0, range: -10.0 to 10.0) - `adapter_name`: The name of the adapter to use (default: "default") - **Outputs:** - `pipeline`: The modified pipeline - **Category:** Diffusers/Lora ## Usage Examples The nodes are designed to work together in a pipeline: 1. Use `DiffusersPipeline` to load your model (select appropriate pipeline type) 2. Encode your prompts using `DiffusersTextEncode` (with optional image for edit pipelines) 3. Apply LoRAs if desired using the LoRA utility nodes: - Load with `DiffusersLoadLoraOnly` - Modify with `DiffusersLoraLayersOperation` (optional) - Apply to pipeline with `DiffusersMergeLoraToPipeline` 4. Generate images using `DiffusersSampling` with your conditioning and parameters For image editing workflows, use `LongCatImageEditPipeline` with an input image to the sampling node. ## Contact - **Twitter**: [@Lrzjason](https://twitter.com/Lrzjason) - **Email**: lrzjason@gmail.com - **QQ Group**: 866612947 - **Wechatid**: fkdeai - **Civitai**: [xiaozhijason](https://civitai.com/user/xiaozhijason) ## Changelog - **Jan 16, 2026**: Added support for GLM Image models (both text-to-image and image-to-image) with dedicated installation instructions - **Earlier**: Initial release with LongCat 6B Image and LongCat 6B Image Edit support ## Sponsors me for more open source projects:

Buy me a coffee:

WeChat:

Diffusers Load LoRA Only

Inputs (1)

Outputs (1)