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-ReCaVSR > ReCaVSR Model Loader

ComfyUI Node: ReCaVSR Model Loader

Class Name

ReCaVSRLoader

Category
ReCaVSR
Author
PedroMarinhoDev (Account age: 50days)
Extension
ComfyUI-ReCaVSR
Latest Updated
2026-11-12
Github Stars
0.03K

How to Install ComfyUI-ReCaVSR

Install this extension via the ComfyUI Manager by searching for ComfyUI-ReCaVSR
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter ComfyUI-ReCaVSR 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

ReCaVSR Model Loader Description

Loads the ReCaVSR transformer, text embeddings, and FlashDecoder for video super-resolution.

ReCaVSR Model Loader:

ReCaVSRLoader, displayed in ComfyUI as “ReCaVSR Model Loader,” prepares the model components needed for ReCaVSR video super-resolution, which increases video resolution while reconstructing visual detail. ReCaVSR uses a Wan2.2-based, one-step streaming diffusion transformer: rather than repeatedly refining each result through many diffusion steps, it processes video in successive blocks and reuses information from earlier processing. The loader brings together this transformer, a precomputed text-conditioning embedding, and FlashDecoder, a lightweight streaming decoder that converts the model’s internal video representation into visible frames. It supports the standard bfloat16 transformer and a compatible INT8 ConvRot version, offering a choice between the reference weight format and a smaller, potentially faster alternative. The node prepares these components for a separate upscaling stage; it does not resize or generate video by itself. Loading is integrated with ComfyUI’s model management, with components initially loaded into RAM and moved to the execution device when first used.

ReCaVSR Model Loader Input Parameters:

transformer

Required dropdown selection identifying the ReCaVSR transformer weights to load. Its options are the files returned by the extension’s model-file discovery function; the supplied code does not enumerate a fixed list of filenames. The default is transformer.safetensors when that filename appears in the available options; otherwise, the schema specifies no explicit default. Minimum and maximum values do not apply because this is a file selection, not a numeric setting. Choose the standard bfloat16 transformer for the reference format, or a compatible INT8 ConvRot file produced by tools/convert_int8.py for reduced weight size and potentially faster execution. The node’s tooltip describes the INT8 version as half the size and 1.5 times faster, with 45.5 dB agreement against bfloat16; these are stated comparison figures, not guaranteed performance on every system. That INT8 option requires an NVIDIA GPU with compute capability 7.5 or newer. Your selection also determines where the loader looks for its companion files: prompt.safetensors and flashdecoder.safetensors must be in the same directory as the selected transformer. Neither companion file is selected separately, and execution stops if either is missing.

ReCaVSR Model Loader Output Parameters:

model

A custom RECAVSR model object containing the prepared transformer, FlashDecoder, precomputed positive prompt embedding, and detected transformer precision information. This output represents the processing components needed by a compatible ReCaVSR upscaling node, rather than an image, video, or ordinary ComfyUI diffusion-model output. Connect it to the model input of ReCaVSRUpscale to use the loaded components for video super-resolution. The prompt embedding supplies the fixed text conditioning used by ReCaVSR, while the transformer performs the core enhancement processing and FlashDecoder supports streaming conversion into visible frames. Receiving this output means the loader has completed its preparation, but it does not mean that upscaling has run or that all components are already resident on the GPU; the loading code indicates that transfer to the execution device occurs on first use.

ReCaVSR Model Loader Usage Tips:

  • Place the ReCaVSR release files in ComfyUI/models/recavsr. Keep transformer.safetensors, prompt.safetensors, and flashdecoder.safetensors together; the loader resolves the two companion files relative to whichever transformer you select.
  • Start with the standard transformer.safetensors when establishing a working workflow. Once it loads and runs successfully, consider a compatible INT8 ConvRot version if you need smaller transformer weights or improved execution speed.
  • Use the INT8 option only with an NVIDIA GPU of compute capability 7.5 or newer, as specified by the node’s tooltip. Create that version with the extension’s tools/convert_int8.py rather than assuming any INT8 model file will be compatible.
  • If you keep multiple transformer variants in different folders, place the matching companion files alongside each selected variant. Having those files elsewhere in the model directory does not satisfy the loader’s same-folder check.
  • Connect model to ReCaVSRUpscale and configure video processing there. This loader has no controls for resolution, seed, custom prompt text, or decoder selection.
  • Allow time and available system RAM for loading, as well as sufficient device memory for the subsequent upscaling run. INT8 reduces transformer weight storage, but it does not establish that the entire workflow’s memory usage will be halved.

ReCaVSR Model Loader Common Errors and Solutions:

prompt.safetensors is missing next to …; download kopper/ReCaVSR into models/recavsr

  • Explanation: This is an explicit error raised when the selected transformer’s directory does not contain prompt.safetensors. The loader needs this file’s precomputed positive prompt embedding.
  • Solution: Download prompt.safetensors from the matching ReCaVSR release and place it directly beside the selected transformer. Check that the filename is exact and that the file is not inside an additional nested folder.

flashdecoder.safetensors is missing next to …; download kopper/ReCaVSR into models/recavsr

  • Explanation: This is an explicit error raised when flashdecoder.safetensors is absent from the selected transformer’s directory. The loader always prepares FlashDecoder, regardless of which decoding method you may later choose in the upscaling node.
  • Solution: Download the matching flashdecoder.safetensors and place it in the same directory as the selected transformer. A copy in a different model folder will not pass this check.

Selected transformer file cannot be found

  • Explanation: The transformer path is resolved before the companion files are checked. If you moved, renamed, or deleted the selected file, path resolution can fail; the exact error wording depends on ComfyUI.
  • Solution: Confirm that the transformer exists in the extension’s model location, refresh or restart ComfyUI to update the available choices, and select an existing file again.

KeyError: 'positive'

  • Explanation: The loader reads the positive entry from prompt.safetensors. This error can occur if the file loads but does not contain that entry, such as when an unrelated embedding file has been given the expected filename.
  • Solution: Replace it with the original prompt.safetensors supplied for ReCaVSR. Renaming a different prompt or model file does not make it compatible.

Safetensors file-loading error

  • Explanation: A malformed, incomplete, or incorrect weights file can fail when ComfyUI reads it. The supplied code does not define a fixed message for this case; the underlying loading library supplies the details.
  • Solution: Use the error’s file path to identify the affected download, download that file again, and confirm that it is an actual model file rather than a downloaded web page or an unfinished transfer.

Model weight mismatch or incompatible checkpoint

  • Explanation: ReCaVSR expects its specific transformer architecture and matching decoder. A generic Wan2.2 checkpoint or an unrelated quantized file may have incompatible parameter names or tensor sizes. The precise error message depends on where loading or execution detects the mismatch.
  • Solution: Use the released ReCaVSR transformer and companion files. For INT8, use the conversion method identified by the extension rather than a checkpoint prepared for another model or quantization system.

INT8 execution fails on unsupported hardware

  • Explanation: The node’s tooltip requires an NVIDIA GPU with compute capability 7.5 or newer for the INT8 ConvRot option. Unsupported hardware or unavailable quantized operations may produce an error during loading or first execution; no exact message is specified in the supplied code.
  • Solution: Check your GPU’s compute capability before selecting INT8. If your hardware does not meet that requirement, switch to the standard bfloat16 transformer and verify its compatibility with your execution environment separately.

Out of memory during loading or first execution

  • Explanation: The loader initially brings weights into system RAM, and model components move to the execution device on first use. Insufficient RAM can interrupt loading, while insufficient GPU memory can cause a later device-allocation failure, often reported as CUDA out of memory on NVIDIA systems.
  • Solution: Close other memory-intensive applications and unload unused models. On supported hardware, consider the INT8 transformer to reduce transformer weight memory. If the failure occurs during upscaling rather than loading, review the downstream node’s video dimensions and available memory-saving settings.

ReCaVSR Model Loader Related Nodes

Go back to the extension to check out more related nodes.
ComfyUI-ReCaVSR
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.

ReCaVSR Model Loader