Skip to content

Sandbox Mode

Sandbox Mode is the default workspace: pick an operation, feed it sources, generate, review. This page tours the layout, then walks through your first generations.

The layout

Across the top: the burger menu, your project name, the centered mode nav (Sandbox Mode, Asset Mode, Batches, Retarget, Workflows, Tracking, Storage), the Assistant toggle, and the Settings gear.

Below that, three columns: the Pipelines panel on the left, the main viewer in the center, and the Inspector on the right, all sitting above a bottom panel with Files and Jobs tabs. When you open the Assistant, it docks on the left.

The Pipelines panel

Finding an operation

Modality tabs run across the top: Text, Image, 3D, Audio, Video. Under them, the Operations browser is a collapsible folder tree grouped by what operations do. Top-level folders include "Text to Image", "Edit Image", "Remove Background", "Depth Map", "Upscale", "Segment", "Restyle", "Reframe", "Virtual Try-On", "Generate 3D Model", "Edit 3D Model", "Generate Splat", "Multi View", "Rig & Animate", "Animate Character", "Motion", "Text to Speech", "Speech to Speech", "Sound Effects", "Music", "Text to Video", "Image to Video", "Reference to Video", "Edit Video", "Extend Video", "Add Audio", "Talking Avatar", "Text Generation", "Transcribe", and "Translate".

The search field ("Search all modalities...") sweeps every modality at once, not just the current tab. The filter popover narrows the tree by minimum star rating, favorites only, or provider (Fal, Fal (Direct), Google, OpenAI, OpenRouter, Local (Plugins), Local (ONNX)).

Right-click any operation to give it a star Rating or mark it a Favorite. Rate the ones that work for your style and the filter popover becomes your personal shortlist.

Reading an operation

Select an operation and you'll see its title, an execution chip, the endpoint id, and a description. The chip tells you where it runs: a teal "Runs locally" chip, a grey "Runs on Fal" chip, or an amber variant meaning it runs on Fal now but a local model is available to download. Clicking the amber chip opens Settings so you can grab the model. See Connect providers.

Source assets and references

Operations that take inputs show numbered slots above the parameters: small thumbnail chips plus a dashed tile for adding, swapping, or dropping.

  • Drag an asset from the Files panel onto the slots to append or replace.
  • Click the dashed tile to open the ReferencePicker, a searchable grid where you choose from your project's assets.
  • Drag chips to reorder, or use Alt+Left / Alt+Right with a chip selected. Del removes one.
  • On multi-reference slots, each chip gets a token badge like @Image3. Cite that token in your prompt to tell the model which reference you mean.

Parameters and presets

Below the slots, the form shows the operation's parameters: text fields, numbers, checkboxes, dropdowns, JSON fields, model pickers. Whatever the operation exposes.

The Presets box saves a full parameter setup under a name. Type into "New preset name..." and save, load one from the "Load preset..." select, or delete one you've outgrown. Presets export and import as .kitlypreset files, so you can share a dialed-in look with a teammate. See File formats.

Generate your first image

  1. Open the Image tab and pick an operation under "Text to Image".
  2. Write a prompt. Something concrete: "orthographic model sheet of a goblin warrior, front and side".
  3. Click Generate.

The job appears inline with a cancel X, and in the Jobs panel. When it completes, the result shows in the viewer, registered into images/generated. While jobs run, the button reads "Generate (N running)"; you can keep queuing more.

If you select something else mid-flight, finished results announce themselves with a "N new outputs" strip. Click it to view them.

Sweeps

Next to Generate, the sweep select (x1, x2, x4, x8) submits that many generations at once, each with a distinct random seed. Same prompt, same settings, different dice. It's the fastest way to get a spread of options to pick from. For sweeping actual parameter values across many rows, use Batches.

Your first 3D, audio, and video

The flow is identical in every modality; only the sources change:

  • 3D: pick an operation under "Generate 3D Model", feed it your best concept image from the reference slots, generate. The result opens in the 3D viewer.
  • Audio: "Text to Speech", "Sound Effects", or "Music" work from a prompt alone.
  • Video: "Image to Video" animates a still; "Text to Video" starts from nothing but the prompt.

The viewer

The center viewer adapts to whatever's selected, with a contextual toolbar.

  • Channels: a switcher flips between "Original" and any depth, masks, splats, subtitles, audio, or animation channels the asset carries.
  • Images: pan and zoom (slider, percentage readout, "Fit to window", "100%"). View modes 2D, 3D (a relief built from the color and depth pair), and Fly (walk through the image in a small environment, with options for facing mode, alpha handling, copies, field of view, environment, grid, and fog). Depth controls offer Invert Depth, threshold and extent sliders, and "Export baked GLB". An ops bar underneath gives Crop (with aspect lock and ratio; Enter applies) and Pad....
  • Geometry and cameras: a GLB viewer. Face mocap results play their captured texture sequence alongside the source footage.
  • Animations: BVH clips play on a paired or default rig, with the source footage and pose overlay when the clip came from mocap. The transport has play and pause, a clip dropdown, scrubbing, and a settings popover (loop mode Repeat, Once, or Ping-Pong, speed 0.1x to 4x, smoothing, autoplay, root motion, skeleton, wireframe).
  • Audio and video: a media player with display segments Play, Wave, Spectro, and Freq, left and right level meters, a subtitle track dropdown, and Transcribe and Translate buttons. Video assets add a full editing toolbar, covered in Video tools.
  • Text: markdown rendering with a raw toggle.
  • Anything visible: Copy to Clipboard and Save Screenshot capture the current view.

The Inspector

With an asset selected, the Inspector shows: Asset (role, path, "Reveal in Explorer"), Channels (view and quick-generate rows, see Assets and files), Prompt and Model (what made it, each with a button to copy it back into the Pipelines panel), Settings (all recorded inputs), Lineage (creation time and content fingerprint), and Operations, a list of operations that accept this asset's kind. Click one and it loads into Pipelines with the asset already set as the source. That last part is the iteration loop: generate an image, click "Image to Video" in its Operations list, generate again.

The Files panel

The Assets tab shows registered assets as draggable cards, with kind filter tabs, a thumbnail size slider, and search. Ctrl or Shift click to multi-select; a bar appears with Add to Kitly Asset..., Delete, and Clear selection. Ctrl+V pastes a clipboard image or text into the project, and OS drag-drop imports files.

Every asset can carry a star rating and a favorite heart. Hover a card to set them, or use the Inspector's Asset section for the selected asset. The toolbar's sort menu orders by Newest, Oldest, Rating, Favorites first, Name, or Kind, and the heart toggle filters to favorites only. Rating something well is how you find it again next week.

A Grid and Table toggle switches the layout. The table adds sortable columns: Name, Kind, Modality, Model (the operation that generated the asset, or "Imported"), Kitly Asset (the container it belongs to, if any; click it to jump there), Rating, Fav, and Created. Click a column header to sort, click again to flip the direction. Selection, drag, and the right-click menu work the same in both layouts.

The Folders tab browses the raw project folders from the "Project" breadcrumb, shows file sizes, accepts drops, and offers Duplicate, Move, and Delete on selections.

Right-click an asset for the full menu: "Reveal in Explorer", "Open With" (your configured external apps, see Settings), "Discuss in Chat" for images, Copy, Paste, Duplicate, Rename, "Move...", Delete, "Create Kitly Asset...", "Set as Channel...", "Create Mocap..." on images and video, "Generate Image Sequence..." and "Re-encode..." on video, "Copy Path", "Copy As Reference", and "Copy to Clipboard" for images and text.

Where results go from here

  • Promote good results into a container: Asset Mode.
  • Vary parameters across a grid: Batches.
  • Chain operations into a repeatable graph: Workflows.