Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > ComfyUI-LTX2.5-MSR > ComfyUI-LTX2.5-MSR IC-LoRA Loader

ComfyUI Node: ComfyUI-LTX2.5-MSR IC-LoRA Loader

Class Name

ComfyUILTX25MSRICLoRALoader

Category
ComfyUI-LTX2.5-MSR
Author
liconstudio (Account age: 478days)
Extension
ComfyUI-LTX2.5-MSR
Latest Updated
2026-09-24
Github Stars
0.04K

How to Install ComfyUI-LTX2.5-MSR

Install this extension via the ComfyUI Manager by searching for ComfyUI-LTX2.5-MSR
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-LTX2.5-MSR in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

ComfyUI-LTX2.5-MSR IC-LoRA Loader Description

Loads LTX MSR or AVref LoRAs and extracts reference-slot embeddings for guided generation.

ComfyUI-LTX2.5-MSR IC-LoRA Loader:

The ComfyUI-LTX2.5-MSR IC-LoRA Loader prepares a compatible LTX MSR or AVref LoRA for reference-guided generation. A LoRA is a compact collection of learned adjustments that changes how a diffusion model generates content. In addition to loading these adjustments, this node extracts learned reference-slot embeddings: stored information that helps the reference-guidance system distinguish and organize reference material. This makes it more specialized than a general-purpose LoRA loader, because it preserves reference-related information needed by the MSR workflow rather than treating the checkpoint only as a set of model adjustments. It also detects AVref checkpoints through their audio-slot data or metadata and selects a dedicated loading path that supports sparse audio time windows. The node prepares these capabilities for use in a larger workflow; it does not generate images, video, or audio by itself.

ComfyUI-LTX2.5-MSR IC-LoRA Loader Input Parameters:

model

A required MODEL connection containing the diffusion model to which the selected LoRA will be applied. You should supply a model compatible with the LTX MSR or AVref checkpoint you intend to use, because LoRA adjustments are designed for particular model structures and are not interchangeable across unrelated model families. This connection supplies the generation model, not a reference image, audio file, or text encoder. The standard MSR loading path applies adjustments only to the diffusion model and does not load or modify a CLIP text encoder. There is no numeric range, selectable option list, or default model defined for this parameter.

lora_name

A required COMBO selection identifying the LoRA checkpoint to load. Its options come from ComfyUI’s registered loras file list, so the available choices depend on the files installed in your configured LoRA locations, commonly ComfyUI/models/loras. The source does not define a fixed list or an explicit default selection. Choose a compatible LTX MSR or AVref checkpoint: the node reads both its weights and metadata, separates ordinary LoRA adjustments from reference-slot information, and determines whether to use the standard MSR path or the AVref path. Audio-slot data, the metadata flag reference_audio_slot_embedding_enabled, or a truthy reference_audio_conditioning metadata value triggers the AVref path. An ordinary style LoRA should not be assumed compatible simply because it appears in this list.

strength_model

A required FLOAT controlling the strength passed to the LoRA-loading process. The default is 1.0, the minimum is -100.0, the maximum is 100.0, and the adjustment step is 0.01. In the standard MSR path, 1.0 applies the LoRA at its unscaled strength, values between 0.0 and 1.0 reduce its contribution, and values above 1.0 amplify it. Negative values reverse the direction of the LoRA adjustments rather than providing a straightforward “less influence” setting, and extreme magnitudes can produce poor or unstable results. At exactly 0.0, the standard MSR path returns the incoming model without applying the ordinary LoRA weights, but still reads the checkpoint, validates the reference-slot information, and prepares the reference parameters. Consequently, zero strength is not a complete bypass of checkpoint processing or reference guidance. AVref loading receives the same strength value, but its detailed zero-strength behavior is not shown in the provided source.

ComfyUI-LTX2.5-MSR IC-LoRA Loader Output Parameters:

model

A MODEL connection containing the model returned by the selected loading path. For a standard MSR checkpoint at nonzero strength, this is the diffusion model with the ordinary LoRA adjustments applied at the requested strength; at zero strength, it is the incoming model unchanged by those adjustments. For an AVref checkpoint, it is the model returned by the dedicated AVref loader. Connect this output to the downstream model-processing or sampling part of your workflow so generation uses the prepared model. This output is not a generated image, video, audio clip, or text encoder.

msr_parameters

An LTX_MSR_REFERENCE_PARAMETERS connection carrying the reference-specific information prepared from the checkpoint. This is a specialized workflow data package, not an image or a numeric strength value, and it is intended for nodes that accept this custom type. In the standard MSR path, it contains the learned reference-slot tensors as slot_state, a copy of the checkpoint’s metadata, the selected lora_name, and reference scaling settings. The reference_downscale_factor is read from metadata, defaults to 1 when absent, and is rounded to an integer with a minimum of 1. The reference_temporal_scale_factor is always set to 1 in this path to preserve ComfyUI’s established guide-coordinate behavior, even if the training metadata records another temporal scale. For AVref checkpoints, this output is prepared by the dedicated AVref loader; its complete internal structure is not included in the supplied source. Keep this connection paired with the model prepared from the same checkpoint so downstream reference processing receives the corresponding learned information.

ComfyUI-LTX2.5-MSR IC-LoRA Loader Usage Tips:

  • Start with strength_model at 1.0 unless your checkpoint’s documentation recommends another value. Compare small changes, such as 0.8 and 1.0, while keeping your prompt, seed, and other settings fixed so you can judge the effect of the strength adjustment.
  • Select a checkpoint specifically intended for LTX MSR or AVref workflows. Being listed in lora_name only means ComfyUI can find the file; it does not confirm that the file contains the required reference-slot information or matches your model.
  • Use both outputs in the appropriate parts of your workflow. Send model onward through the model or sampling path and connect msr_parameters to a compatible reference-guidance input, such as the corresponding input on the companion Multi-Reference Guide where available.
  • Treat strength_model as LoRA-adjustment strength rather than a universal reference-influence control. In the standard MSR path, reducing it to zero does not remove the extracted reference parameters or stop them from being used by downstream nodes.
  • Let checkpoint metadata provide the reference downscale setting. There is no control for that setting on this node, and the standard MSR path deliberately fixes the temporal scale factor at 1 for compatibility; do not assume a different temporal scale recorded during training will be used here.
  • For audio-reference workflows, use an appropriate AVref checkpoint and compatible downstream audio-reference processing. The node detects AVref support from checkpoint contents and metadata; there is no separate audio-mode switch on this interface.
  • Check the ComfyUI console when diagnosing standard MSR loading. Successful loading logs the checkpoint name, the number of learned reference-slot tensors, and reference settings. The temporal-scale message marked as a compatibility default is informational, not an error.

ComfyUI-LTX2.5-MSR IC-LoRA Loader Common Errors and Solutions:

LoRA file not found or selected file cannot be resolved

  • Explanation: The node resolves lora_name through ComfyUI’s registered LoRA folders. If the file was moved, deleted, renamed, or is outside those locations, loading can fail before its contents are read. The exact exception wording depends on ComfyUI’s file-path handling.
  • Solution: Put the checkpoint in a configured LoRA folder, refresh the file list or restart ComfyUI if necessary, and select the file again. If you use additional model paths, verify that they are configured for LoRA files.

Checkpoint cannot be read or fails safe loading

  • Explanation: The node reads the checkpoint with safe loading enabled. A damaged download, an unsupported file format, insufficient file permissions, or checkpoint contents incompatible with the safe loader can prevent loading. The supplied class does not define a single fixed error message for these failures.
  • Solution: Confirm that the file is a supported checkpoint rather than an archive or download-page file, check that ComfyUI can read it, and download a fresh copy from a trusted source if necessary. Use a checkpoint release intended for this workflow rather than disabling loading safeguards.

Reference-slot validation fails

  • Explanation: The standard MSR path validates the extracted reference-slot information before applying the LoRA. Missing or incompatible slot tensors or metadata may therefore stop loading, even if the file is a valid ordinary LoRA. The validation helper’s exact error messages are not included in the supplied source.
  • Solution: Verify that you selected an LTX MSR reference checkpoint rather than a general-purpose LoRA. Check the checkpoint’s installation instructions and required extension version, and replace incomplete or incompatible files with the appropriate release.

KeyError: 'net.2.bias'

  • Explanation: The standard MSR path directly accesses net.2.bias in the reference-slot state when reporting the slot output dimension. If this error appears, that expected tensor is missing at the point of access. Validation may catch such a problem earlier, so this exact exception is not guaranteed for every malformed checkpoint.
  • Solution: Check that the checkpoint uses the reference-slot format supported by your installed extension. Obtain a compatible checkpoint or update the extension according to its release instructions; avoid manually renaming tensors, because matching names alone does not make their contents compatible.

LoRA weights do not match the model, or some weights are not applied

  • Explanation: A LoRA must match the model structure it was trained for. An incompatible combination can cause loading errors or unmatched-weight warnings, depending on the underlying loader. Successful file reading alone does not prove that the adjustments were applied correctly.
  • Solution: Confirm the checkpoint’s required base model and model version, then connect the matching model. Review console warnings before troubleshooting prompts or sampler settings, and first test the reference LoRA without unrelated model modifications.

Invalid reference_downscale_factor metadata

  • Explanation: In the standard MSR path, reference_downscale_factor is converted to a floating-point number, rounded, and clamped to at least 1. Metadata that cannot be converted to a usable number can cause a conversion or rounding exception. This value is stored in the checkpoint; it is not an input control on this node.
  • Solution: Use an intact checkpoint with valid metadata or obtain a corrected release from its publisher. Changing strength_model will not repair malformed metadata because the standard path still processes reference parameters at zero strength.

AVref checkpoint fails in the dedicated loading path

  • Explanation: Audio-slot data or audio-related metadata sends the checkpoint to the AVref loader. Errors after that point may involve AVref-specific compatibility or checkpoint contents, but the helper implementation and its exact error strings are not provided here.
  • Solution: Check that the file is a complete AVref checkpoint supported by your installed extension and that you are using its recommended base model. Use the full console traceback to identify the failing component, and consult the matching checkpoint and extension instructions.

msr_parameters cannot connect to a downstream input

  • Explanation: This output uses the custom type LTX_MSR_REFERENCE_PARAMETERS. A standard MODEL, CONDITIONING, IMAGE, or other unrelated input cannot accept it. Missing or incompatible companion nodes can also prevent you from building the intended reference-guidance connection.
  • Solution: Connect it only to an input explicitly accepting LTX_MSR_REFERENCE_PARAMETERS. Make sure the companion reference-guidance nodes are installed and loaded from a compatible version of the extension, then restart ComfyUI if you updated the installation.

ComfyUI-LTX2.5-MSR IC-LoRA Loader Related Nodes

Go back to the extension to check out more related nodes.
ComfyUI-LTX2.5-MSR
RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

ComfyUI-LTX2.5-MSR IC-LoRA Loader