Load Image From Output With Workflow Ovum

Core ComfyUI's Load Image gives you pixels. This node gives you pixels plus the recipe: it loads an image from your output folder and extracts the embedded prompt and workflow metadata stored in the PNG, so you can feed both the image and its provenance back into the graph.

The name is a mouthful, but the idea is simple and genuinely useful: ComfyUI stamps the workflow JSON into every saved PNG, and this node is built to round-trip it. Load a previously generated image, and its PROMPT&WORKFLOW output reconstructs the graph that made it - which is the closest thing to "undo" a batch of generations has ever had. Drag an old output in, keep the image, tweak the prompt, re-run. That's the workflow.

Note the EXPERIMENTAL = True flag in the source - it works, but don't be surprised if the edge cases (weird metadata, files saved by other tools) need a nudge.

Inputs and outputs

• image - a COMBO with upload support, pointing at the output folder (image_folder: "output"), with a refresh button and a control_after_refresh option so it picks up newly saved files. Pick a file from the dropdown or upload one.
• output_subdir - not here; that's the subdirectory variant's job.

Five outputs:

• IMAGE - the loaded image, ready for a sampler, upscaler, or VAE.
• MASK - the image's alpha as a mask, if it has one.
• FILEPATH (STRING) - the resolved path, handy for filename-based logic.
• PROMPT&WORKFLOW (DICT) - the embedded prompt and workflow metadata parsed out of the file.
• IMAGE_EX (DICT) - the pack's image-plus-metadata bundle, for wiring into other IMAGE_EX-aware nodes (LiveCrop, the context nodes).

How it differs from core Load Image

Core Load Image can also read embedded workflow if you dig for it; this node makes it a first-class output and adds the IMAGE_EX bundle plus the output-folder remote route. The practical difference: you're browsing your generated output (not the input folder), and the metadata comes out in a usable DICT rather than being a side quest. It also mirrors the core node's LOADED_PATH hidden plumbing, so saving and re-loading stays consistent.

Where you'll use it

• Iterating on old generations. Load the last good image, keep its prompt, change one parameter, re-run - the workflow comes with it.
• Batch re-processing. Point it at an output, chain into an upscaler or img2img, and improve a batch of past results with their own metadata intact.
• Workflow archaeology. Someone sends you a PNG; this node reconstructs the graph that made it instead of you rebuilding from memory.

Install and gotchas

Ships in comfy-ovum:

cd ComfyUI/custom_nodes
git clone https://github.com/sfinktah/comfy-ovum

or ComfyUI Manager → comfy-ovum, restart. No models, no keys.

Watch for: images saved by tools that strip metadata will come back with an empty or partial PROMPT&WORKFLOW (not an error - just no recipe), and the node is marked experimental, so keep core Load Image as a fallback if something feels off. It browses the output folder specifically - the input folder is the plain LoadImageWithWorkflowOvum's territory.