🇬🇧 English version (default)
🇫🇷 Lire la version française

Auto Cached Preview

<img width="1892" height="865" alt="image" src="https://github.com/user-attachments/assets/1fbcfad3-0d67-4f5f-b337-161f28284d8d" />

A custom node for ComfyUI that allows you to cache an image and its mask to avoid recalculating upstream steps in your workflow.

Ideal for single-image creation workflows where you want to adjust final parameters (upscale, filters, color correction, inpainting) without having to rerun your KSamplers.

✨ Features

• Clipboard image paste: Paste an image directly from your system clipboard into the node cache using the Paste image button.
• Instant cache clearing: Remove the cached image and clear the node preview with the Clear cache button.
• Automatic caching: Saves the last received image and its mask to the disk.
• Smart fallback: If you disconnect the node's image input, it instantly reloads the last cached image.
• Automatic mask management: Automatically extracts the mask from the image's Alpha channel if it exists. Otherwise, generates a default empty mask.
• Startup cleanup: The temporary cache folder is automatically cleared every time ComfyUI is launched to prevent filling up your hard drive.
• Metadata preservation (PNG Info): Generated preview images retain the prompt and the complete workflow. You can still drag and drop the preview into ComfyUI to reload your workspace.

---

🛠️ Usage

The node is located in the category: orion4D_image/preview under the name Auto Cached Preview (Image + Mask).

1. Initial generation: Connect the output of your KSampler (or VAE Decode) to the image input of the node. Run the generation. The image is displayed and cached.
2. Using the cache: Disconnect the incoming link to the image input. Ensure the fallback_to_cache option is set to True.
3. Real-time adjustments: Restart your workflow (Queue Prompt). The node will instantly load the image from the disk. You can continue working on downstream nodes without recalculating the original image.

Inputs

• image (optional): The incoming image tensor to be cached.
• mask (optional): The mask tensor. If not provided, the node deduces it from the image or creates a default one.
• fallback_to_cache (BOOLEAN): If enabled (True), the node will use the cached image if the image input is disconnected.

Outputs

• image: The image (either the input one or the one retrieved from the cache).
• mask: The associated mask.

---

📁 Technical details

Cache files (.pt) and previews (.png) are saved in the ComfyUI temporary folder, under the AutoCachedPreview subfolder. This folder is purged at every startup via the __init__.py script to ensure healthy disk space management.

---

<div align="center">

Made with ❤️ for the ComfyUI community · Orion4D

</div>