Prompt Gallery

LoRA collections get big fast, and filenames lie to you. hair_style_final_v3 tells you nothing until you've rendered it ten times. Prompt Gallery fixes that by rendering your LoRAs, embeddings, and saved tag groups as a grid of preview-image tiles right inside the node - click a tile and it joins your prompt as a tag. It's the visual browser the rest of the EreNodes pack feeds into, and it's the reason most people install this pack at all.

What it actually is

Prompt Gallery is the newest node in ComfyUI-EreNodes (added in v2.1), and it's the pack's showpiece. Where the sibling nodes render tags as text pills, this one renders them as thumbnails in a grid, 100×100 by default. You browse with your eyes, click the ones you want, and the node emits the resulting prompt as a STRING. It handles three content types: LoRAs, embeddings, and Tag Groups (the pack's saved prompt collections).

How it works

The mechanism is a nice split between frontend and backend. The backend walks ComfyUI's folder_paths for your loras and embeddings directories - which means it respects extra_model_paths.yaml, so models on secondary drives show up too. Previews come from a /erenodes/view endpoint that looks for an image sitting next to the model file: either <name>.<ext> or <name>.preview.<ext> (so mylora.safetensors wants mylora.png or mylora.preview.png). No image file? The endpoint returns an empty response and the tile renders blank - no error spam, just a gap.

The frontend then draws each tile with a lazy-loaded <img>, sized by two node properties (_tagImageWidth, _tagImageHeight, both default 100). Right-click a tile for the quick-edit menu (strength control, toggling a LoRA's trigger words); click-drag changes strength. The output is the same tag string format as the rest of the pack - plain text with <lora:name:strength> markup where LoRAs are involved.

Inputs and outputs

• text - the multiline STRING that holds your selected tags/LoRAs. It's the node's working list, edited visually via the tiles.
• prefix (optional) - STRING prepended to the output.
• Output - one STRING. Feed it to CLIP Text Encode, or to the pack's Prompt to LoRA Stack node if you want the <lora:...> tags turned into a real LORA_STACK for stack-compatible loaders (Efficiency Nodes, ComfyRoll, LoRA Manager).

Installing it

ComfyUI Manager: search "EreNodes", install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Erehr/ComfyUI-EreNodes.git

Restart ComfyUI and the nodes appear under EreNodes. No heavy Python deps - safetensors (already bundled with ComfyUI) is the only declared one; the real work is frontend JS.

Gotchas

The known pain point is real and documented in the release thread: a gallery with dozens of LoRAs can tank your framerate while it's open (one user reported 130→4 fps). The author's fix, and the right habit anyway: use ComfyUI-Lora-Manager to fetch small 480px WebP previews into the LoRA folders - small images are exactly what this node wants, and the .preview naming it looks for is what Lora-Manager writes. Collapse the node when you're not browsing. And remember this is a prompt builder, not a LoRA loader - if you click tiles and then feed the raw string to a text encoder, the <lora:...> markup gets encoded as literal text. Route through Prompt to LoRA Stack or a prompt-reading loader like LoRA Tag Loader instead.