Embeddr Find Similar (Text)
Search your image library with a sentence, not a filename
- images
- embeddr_ids
Find Similar (Text) is the semantic-search node: type a prompt - not a filename, an actual description - and it searches your Embeddr library for images that match the meaning. "neon city street at night, rain" returns your rainy-night shots, even if none of them have a filename that says so. It's the text-to-image search you wish every folder-based library had, and it's this pack's best discovery feature.
If you've built up a large library of generations, this is the node that finally makes it searchable by vibe instead of by memory. The output is ready-to-use images, so you can round up a set of reference shots for a style, or find every previous attempt in a series and feed them into an img2img workflow.
The inputs that matter
prompt- a multiline text field. This is your query, and it's the whole game: the better you describe the content, the better the results.library/collection- two combo dropdowns, both defaulting to"All". The node fetches your libraries and collections live from the server when it's created, so you'll see real names like"42: my-style-refs"in the dropdown (that's theID: Labelformat the backend returns). Pick one to scope the search; the node parses the ID out of the"ID: Label"string and sends it aslibrary_idorcollection_id. Leave both on All to search everything.limit- how many results, 1–50, default 5.
Mechanically it's a GET to /api/v1/images on your Embeddr server with q, limit, and the optional scope IDs - your backend does the embedding/retrieval, this node does none of it locally.
Outputs
images- the matching images as anIMAGElist.embeddr_ids- note the name: it'sembeddr_ids, notartifact_ids. Same idea - the IDs of the results, in order - but the author named this output differently, so don't go hunting forartifact_idswhen you wire this one up. Feed an ID into Find Similar To Artifact to chain a text query into an image query.
A search that finds nothing returns an empty 64×64 image and an empty ID rather than crashing, so an all-black result means "no matches," not "broken." Before you blame the node, make sure your Embeddr backend is actually running (embeddr serve) - if it's down at graph-build time, the dropdowns won't populate with anything beyond "All," which is your first clue.
Install
The pack's shared story, briefly: install embeddr-cli, run embeddr serve, get the pack via ComfyUI Manager (search "embeddr") or a release download into custom_nodes, restart ComfyUI. Declared dependency is just requests. This is a bridge node - your prompt goes to your server on localhost, not a cloud, which is the whole point of the self-hosted setup.
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| prompt | STRING | — | |
| library | COMBO | All | 1 options: All |
| collection | COMBO | All | 1 options: All |
| limit | INT | 51–50 | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| images | IMAGE | — |
| embeddr_ids | STRING | — |