Draggen Extract Images

Draggen Extract Images is where a moodboard stops being a document and becomes pixels you can actually feed the graph. Give it the moodboard output from either loader and it walks every element, keeps the ones that are images, resolves each to a real file, and hands back a list of IMAGE tensors. This is the node for anything that treats reference material as input - an IP-Adapter-style conditioning pass, an img2img run, ControlNet-style reference, or just dumping the assets so you can eyeball them.

How it resolves files is worth knowing, because it's where the Local and Remote paths quietly diverge. The compositor's image resolver tries local paths first - against the board's base path and its images/ folder - and only falls back to fetching src over HTTP when nothing local matches. So a locally-loaded board never needs the network; a remotely-loaded one with absolute URLs works too. Same node, two transport modes, no settings to flip.

Input and output

One input (moodboard), one output: IMAGE, but as a list, not a batch. That distinction is the practical gotcha, and it's deliberate. Board images come in all sizes, and ComfyUI batches need uniform dimensions, so the node returns individual tensors instead of stacking them. A PreviewImage node handles lists fine - that's exactly what the pack's own example workflow wires up. The moment you want to hand the result to something expecting one batched IMAGE - a sampler, a rebatch node - you'll need to convert the list first, and mismatched sizes will bite you.

When things go wrong

Two failure modes worth knowing, both visible in the code:

• If the board has zero image elements, the node returns a single blank 64×64 black tensor. Your graph won't crash, but downstream nodes get garbage unless you branch on it.
• If an individual image fails to load (bad path, dead URL), that element becomes a red 100×100 placeholder instead of raising an error. It's the author's way of not killing the whole run - which also means you should actually look at the preview before trusting the batch.

Installing

Install is the pack-wide story: ComfyUI Manager → search ComfyUI-Draggen, or clone into custom_nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/kadima-tech/comfyui-draggen

then restart. No extra dependencies - just requests and Pillow, which ComfyUI ships anyway.

Where this fits: a moodboard is essentially a curated style reference, the same impulse that sends people toward IP-Adapter or style LoRAs. The difference is curation - you pinned those images in Draggen for a reason, so extracting them into the graph gives you reference material that already passed your editorial filter. It's a small node with a small job, and it does it without fuss.