Files, Flows, Models. Filter All / Style / Object / Character. Search. Sort. Heart the ones you reuse. + Add opens the popup — not another app.
Same list. Two doors
Name, type (Style / Object / Character), drop up to 5 refs, generate the model
Paste a link from Civitai or another model host. Preview fills in. Save to the list
Small frames, odd ratios, baked-in text wreck training. The popup flags them before you spend
Apply all fixes. Then generate
Pick Style, Object, or Character on the Image or Video node. Ask Agent can close the graph with Make it with LoRA. Community models land in the same drawer. You stop prompting the face and start directing the scene
How it locks: look, thing, or face. Style keeps a look (palette, texture, brushwork) across anything you generate. Object keeps one specific thing, like a product or a prop, recognisable from different angles. Character keeps a face and a person consistent across scenes. It is the same trainer underneath; what changes is the kind of reference set you give it and what it holds on to.
A public model URL: Civitai and similar hosts. Paste it, preview it, save it. You are not training; you are filing it. The imported model lives in your library like any other, and you can wire it into a graph.
Browse and import, yes. Training is Studio and Atelier.
Only if you publish it. Private models stay in your drawer, and nobody else can run them or even see them. Publish a character and anyone can run it from the listing, and paid runs cut you in.
