Project Settings

Project Settings configures how the project stores files, how aggressively the Assistant acts, and which services and tools are available. The page autosaves — a status chip shows Saving / Saved, so there's no Save button to press. Settings are organized into tabs; the Access tab appears only if you can manage users.

• Server — files live on the server; works from any browser, no client software
• Local — files live on a local/shared drive; requires LoftDesktop. Configure the drive path (e.g. Z:\Projects\PROJECT0002) and the publish folder name (default "pub")

See the Modes topic for the full comparison.

Show Beta Features is off by default. Turning it on adds the Composer and Flows entries to the sidebar for everyone working on this project. These tools still change without notice, so leave the switch off on a project doing real work.

Who can see this project, and who is on it. The same controls are on the project page (the people icon in the header); this tab is the fuller view.

• Visibility: Everyone — anyone with an account on the site sees the project in their list. This is the default, including for every project created before the setting existed
• Visibility: Assigned team only — the project is hidden from everyone except the team below, plus admins and owners, who always see every project
• Team — the project's crew. Add a person to let them see a restricted project; adding them also grants the access group they need to use the app, if they don't already carry one

The team is the same roster Resource staffs, so someone staffed on the project there is admitted here automatically. Removing a person takes them off this project only — they keep access to any other project they're on, and they keep their account.

Standing preferences every generation starts from. Each is optional; left unset, generation behaves exactly as it did before.

• Preferred Models — which model the Assistant reaches for when a request doesn't name one, set separately for image, video, and image-to-video. A model you name in the request always wins; otherwise the model selected in the chat's model bar comes first, then this preference, then the Assistant's own pick. A preference pointing at a model that's since been disabled falls back automatically rather than failing
• A model that accepts a start frame under a text-to-video card is offered in both the Video and Image to Video slots, and is picked for animating an image just like a dedicated image-to-video model
• Aspect Ratio — the shape of the frame, translated to each model's own format automatically
• Video Resolution — the resolution tier for models that offer one; higher tiers cost more
• Generate Sound — off by default, because native audio bills extra on every provider that offers it. Asking for sound in a specific request still works

Defaults apply only to models that accept the setting, and never override a value the request already specified. Reruns reproduce the original job's settings and ignore these entirely.

Thresholds that trigger a confirmation before the Assistant does something expensive or destructive. Set -1 on any limit to disable it (never ask).

• Video Generation — max videos in one batch before confirming
• Image Generation — max images in one batch before confirming
• Bulk Delete — max items to delete before confirming
• Bulk Update — max items to update before confirming

The Services tab toggles built-in models and registers external services; the Tools tab registers local editor applications. See the Services & Tools topic for details.