Flows

The most direct way in is from the file you want to understand:

1. Open a library or the Files browser and select the generated file(s) you're curious about.
2. Click the flow icon in the toolbar — its tooltip reads "View generation flow".
3. Flow opens and automatically builds the graph by walking backward from that file through its history (you'll briefly see "Loading lineage…").

The result is a left-to-right chain: the inputs on one side, each generation step in the middle, and the file you started from on the other. If a file was made outside Loft (uploaded, not generated), there's no recipe to show and you'll see "No generation history found".

Each box is a node, color-coded by what it represents:

The lines between nodes carry the input's role — for example a start frame, an end frame, or a reference image — so you can see not just which files fed a step but how each one was used. Click any node to open the inspector on the right; double-click a node with a result to open a full-screen preview.

Click an Action node to see its recipe in the inspector: the Type of generation (e.g. Image to Video), the Model used, the full Prompt, the asset it wrote into, and when it ran. This is the read-only "what happened here" view — the starting point for deciding whether to reproduce or change the step.

The Inputs list shows a thumbnail, name, and role for every file the step consumed. Click a row to open that file in the Files browser (an asset input opens its folder), or click the small pinicon on the row to pin it to the Assistant. The same list appears on an Operation node, so a duplicated step's inputs behave the same way.

With an Action (or Queue) node selected, the inspector gives you two ways to redo it. Choose by whether you want the same result or a changed one:

Both leave the original untouched — a redo always writes a newtake, so your history is never overwritten.

To redo a step with adjusted parameters — the common case:

1. Select the Action node that produced the result you want to change.
2. Click Duplicate. A new Operation node appears below it, pre-filled and wired to the same inputs.
3. In that node's inspector, change what you need — pick a different Model from the dropdown, edit the prompt in the prompt builder, or adjust any of the model's parameters. You can also rewire inputs by dragging new connections.
4. Click Generate. The step runs as a new take; the result flows back onto the canvas when it's done.

Nodes on the canvas are stand-ins for real files in the project. Select an image or Asset Item node and the inspector's Location section takes you to it: Open file jumps to the Files browser and opens the file, and Show containing folder lands you in its folder without opening anything. Double-clicking the node still gives you a full-screen preview without leaving the canvas.

Click any image or Asset Item node and the inspector offers a "Generate from this" shortcut — Image to Image, Image to Video, or Upscale — each of which drops a new Operation node wired to that file, ready to prompt and run. This is how you branch a new direction off an existing result rather than redoing an old step.

You can also assemble a pipeline from scratch using the toolbar:

• Add Images — the same picker the generator uses: browse any category or the Pinned view and filter by status; each picked image becomes an Image Source node.
• Add Node — drop in an Operation node from the menu (Text to Image, Text to Video, Image to Image, Image to Video, Video to Video, Upscale).
• Drag from a node's handle to another node to connect an input to a step.
• Save / Save As… — keep the canvas under a name; Open reloads any saved flow.

Select an Operation node to fill in its model, prompt, and parameters, then press Generate. Press Delete orBackspace to remove the selected node or connection.

If the Assistant proposed changes to a step's recipe, that node shows an amber "Changes" pill. Click it to open Review changes, which shows a before/after diff of the prompt and each parameter. Choose Accept to make the edited recipe the new baseline, or Revert to discard the changes and restore the original.

Selecting an image or Asset Item node counts as a file selection everywhere else in the app does — the Assistant sees it for the current turn ("upscale the selected shot" works without naming a file), and its add button in the panel's action shelf pins the node's file into the session context so it stays available across later messages. Select several nodes to hand over several files at once; clicking empty canvas clears the selection again.

For a single file there is a shortcut: with an image or Asset Item node selected, the inspector's Assistant section has an Pin to Assistant button that pins that file straight into the session context. It shows up as a chip in the Assistant panel's tray next to your attached files, and you can remove it there like any other.