Meld Image Loader

ComfyUI's killer feature is that a workflow rides along inside the PNG - drag a generated image back in and the graph rebuilds. Meld Image Loader takes that idea one step further inside the canvas: load an image and it not only gives you the pixels, it parses the embedded generation metadata and hands back the prompt, the settings, and even an attempt at loading the actual model that made it.

It's the "read the DNA" node of ComfyUI-Meld, and it's the natural partner to Meld Save Image: save with Meld, then reload the recipe whenever you want to iterate.

How it works

Drop an image on the node (or pick from the file list) and it tries, in priority order, to parse:

1. ComfyUI workflow JSON - the most detailed, since it contains the whole graph.
2. ComfyUI prompt JSON - used for API-generated images.
3. A1111 info text - for images made in Automatic1111.
4. Defaults - if nothing's found.

From that it extracts seed, steps, cfg, sampler, scheduler, resolution, and prompts. Then it does the slightly ambitious bit: it tries to load the checkpoint recorded in the metadata from your models/checkpoints folder, even attempting a fuzzy name match if the exact filename isn't there. Success isn't guaranteed - if the model file isn't on your disk, MODEL/CLIP/VAE come back empty.

The outputs

• IMAGE - the pixels.
• MODEL / CLIP / VAE - the loaded model trio, or None if the checkpoint isn't present.
• positive / negative - the extracted prompts as strings.
• summary - a text dump of everything it found and decided; feed it into a Show Text node to see the reasoning.
• base_settings - the settings bundle, ready for Meld Settings Unpacker to split into seed/steps/cfg wires.

What it's for

The reuse loop is the headline. Load an old image, and instead of typing its prompt from a screenshot, you've got positive/negative wired straight into a text encoder and base_settings driving a KSampler. Change one thing - the seed, the model, the style - and regenerate. It's also the honest way to check what a stranger's workflow actually used: the summary output lays the parameters out plain.

Installing it

Same pack, same install:

• ComfyUI Manager → search "Meld" → install, or
• comfy node install HappyOnigiri/ComfyUI-Meld, or
• cd ComfyUI/custom_nodes && git clone https://github.com/HappyOnigiri/ComfyUI-Meld.git && cd ComfyUI-Meld && pip install -r requirements.txt then restart.

No model downloads. One honest caveat: metadata extraction works best on images saved by ComfyUI or A1111 that actually carry full workflow JSON - a bare render from a closed-source app or a screenshot won't have much to parse, and you'll get defaults. And since the pack only reads generation info (not EXIF like GPS), it's privacy-considerate by design. If a saved image won't reload its settings, check the summary output to see what it found.