PAL (Previz and Layout) is a spatial intelligence tool for previsualisation and camera layout. It lets you build 3D scenes from text descriptions, position cameras, animate shots, and export the results directly into the LensCowboy pipeline.

PAL is designed for directors, cinematographers, and VFX supervisors who need to block out scenes quickly without deep 3D software expertise. It uses AI to convert natural-language scene descriptions into 3D layouts you can refine and iterate on.

Launch PAL from the main LensCowboy dashboard by selecting a project and clicking Open in PAL. If you have an existing LCBE breakdown, PAL will offer to load its scenes and shots automatically.

You can also open PAL as a standalone workspace and create scenes from scratch without a prior breakdown.

Working on multiple projects at once

Each PAL tab is scoped to its own project. Open two browser tabs from two different LCBE projects and they stay independent — switching projects in one tab doesn't affect the other. Every API call, Drive folder lookup, and shot list belongs to the tab's own project.

The active project name shows in the top bar next to the LENSCOWBOY PAL logo. A small × unlink chip appears next to it whenever PAL is bound to a project — click it to revert that tab to standalone mode (the LCBE project's data is preserved; relaunching from LCBE rebinds instantly).

Each project has its own Drive folder tree (source images, source objects, LUTs, render output, watermarks), so two projects never share files.

The viewport is the main 3D view where you build and review your scene. It has three modes:

Switch between modes using the mode selector in the top toolbar.

Enter a description in the prompt field — a screenplay action line, a shot brief, or plain English:

A dimly lit warehouse interior. Forklift in the foreground.
Two figures stand near a loading dock at the far end. Rain
visible through open bay doors.

PAL's AI reads the description and generates a 3D layout using proxy geometry: rough shapes for walls, floors, objects, and characters placed at estimated positions. This is not final-quality geometry — it is a spatial sketch that you can then refine.

Iterate by updating the prompt, adding detail, or manually adjusting object positions.

Navigation is consistent across Layout, Camera, and Director views:

• Orbit — Alt + left-click drag (Maya nav) or LMB drag (PAL nav)
• Pan — Alt + middle-click drag, or RMB drag in PAL nav
• Dolly / Zoom — mouse wheel, or Alt + right-click drag
• WASD fly — hold RMB, then use WASD to move and mouse to look. Q/E for up/down. Available in all viewports.

All four viewports (Layout, Camera, Director left, Director right) orbit in the same direction and respect the Invert Orbit checkbox in Settings consistently.

In Camera mode, pan and dolly translate along the camera's own axes and preserve orientation — switching Alt+LMB orbit to Alt+MMB pan no longer snaps the camera.

The Director split right panel matches Camera-mode behaviour: middle-click and Alt+MMB both pan the shot camera; Alt+RMB dollies it. Hold RMB without Alt to enter WASD fly through the right pane.

Active camera (VIEW dropdown)

The VIEW dropdown in the camera bar picks which camera the viewport is looking through. Options:

• Native CAM — PAL's default shot camera. Gizmos, keyframes, and DOF all act on this one.
• Imported cameras — any cameras that came in with an FBX or USD import. Cyan icons in the object list.

Selecting an imported camera locks the viewport to its track — orbit, pan, dolly, and wheel are intentionally inert while you're looking through a "filmed" camera, the same way you can't accidentally re-aim a Maya / Blender camera that's locked. Switch back to Native CAM to regain free navigation.

The same dropdown appears in the AI Render and Render & Export dialogs — whichever camera you've activated in the viewport is the one those panels capture from. The three selectors stay in sync.

Camera I/O — .lcam.json round-trip (↑ IMPORT / ↓ EXPORT)

Two small buttons in the camera bar export and re-import the native CAM as a .lcam.json sidecar. The file preserves transform, lens metadata (body / focal length / aperture / focus distance), FOV, and any CAM keyframes on the timeline — a lossless way to move a shot's camera between scenes, share a setup with another artist, or back up before risky edits.

• ↓ EXPORT — saves the current native CAM state (and CAM keyframes if any) to a .lcam.json file.
• ↑ IMPORT — reads a .lcam.json and overwrites the native CAM. If the file contains keyframes you'll be asked whether to replace the existing CAM keyframes on the timeline or apply the static transform only.

Live pass preview (P)

The Camera bar has a passes button (▦) and a P hotkey that cycles the live viewport between five outputs:

• Beauty — full-quality scene with HDRI / sky lighting (default).
• Depth — gradient driven by the camera's clip planes (Clip Near = white, Clip Far = black). The Depth Gamma slider in Settings rolls the midtones up or down.
• Normal — surface-normal pass.
• Alpha — white objects on black background.
• Matte — ID Matte. Each scene object renders as a deterministic unique solid colour — useful as a per-object mask or for downstream compositing / segmentation.

What you see in the preview is exactly what the render dialog will output for that pass — no need to render to verify framing, depth-range, or matte coverage.

Camera HUD & frame guides

Camera mode overlays a few cinematographic guides on the viewport so you can frame shots with awareness of how a real camera body would record them. From outermost to innermost:

• Amber dashed rectangle — the camera body's sensor aspect. The label at its top-left reads CAMERA · <ratio> (e.g. 1.46:1 for Alexa 35 Open Gate, 16:9 for Alexa 35 4.6K 16:9, 3:2 for full-frame stills, etc.). Changes when you pick a different body or recording mode in the camera bar.
• Cyan dashed rectangle — the delivery extraction, drawn inside the sensor frame. Shows how the render resolution (e.g. 1920×1080 = 16:9) crops out of the sensor area. Hidden when the render aspect matches the sensor aspect (no crop needed).
• Inner safe-area lines — traditional title and action safe regions inside the delivery extraction.

Cinema convention: the render extraction is always contained within the sensor — you crop the sensor down, never out. So the cyan rect always sits inside the amber rect, with the relationship fixed by the aspect ratios:

• If render is wider than sensor (e.g. 16:9 render on a 3:2 Open Gate sensor) — cyan uses the full sensor width and gets letterboxed top + bottom
• If render is narrower than sensor (e.g. 4:3 render on a 16:9 sensor) — cyan uses the full sensor height and gets pillarboxed left + right

The relationship stays fixed as you resize the browser; only the absolute size of both rectangles changes with the canvas.

Click any object in the viewport to select it. A selected object displays a transform gizmo:

• Translate (G) — arrow handles, per-axis + plane squares
• Rotate (R) — torus rings per axis
• Scale (S) — cube-tipped handles, centre cube for uniform scale

Drag the coloured handles to constrain movement to a single axis. Exact values can be typed in the Channel Editor (press N or click CH on the timeline).

Each mode hit-tests only its own handles — translate mode never accidentally hits rotate or scale handles. Works in all viewports including Camera view.

Local / World space

Toggle via Insert (fn+Enter on Mac) or the space buttons under shots. Local aligns gizmo axes to the object; World aligns them to world axes.

Reset shortcuts

Ctrl+G position to origin · Ctrl+R rotation to zero · Ctrl+S scale to 1,1,1.

The timeline panel at the bottom of the viewport animates cameras and objects over time. It collapses all tracks into a single row — keyframes from any object appear as dots on the same timeline, colour-coded by channel (translate red, rotate green, scale blue, all channels yellow).

• Scrub the playhead to move through time.
• Position a camera or object, then press K or I to record a keyframe at the current frame.
• In Camera mode with nothing selected, K/I automatically keys the shot camera.
• Move to a different frame, reposition, and key again. PAL interpolates between keyframes (slerp on rotation — gimbal-free).

Interpolation modes

Right-click a keyframe dot to pick: Linear, Flat (hold), Bezier (smooth), Ease In, or Ease Out.

Auto-Key (AK button)

When enabled, moving an object that already has keyframes automatically drops a new keyframe at the current frame on the channels that are already animated. Keeps you from having to press K after every move.

PAL includes a library of 34 built-in proxy types: human figures, vehicles, furniture, architectural elements, props. These are low-polygon stand-ins meant for layout, not final rendering.

Drive asset library

Mount a Drive folder in Project Settings. PAL scans it and lists all compatible files in the asset browser. Drag-to-canvas or double-click to place. Drive-imported assets persist across shot switches.

Supported formats

• .glb / .gltf — recommended. External .bin and textures in sibling subfolders are auto-bundled on load.
• .fbx — materials auto-converted to MeshStandardMaterial so HDRI lighting affects them.
• .obj — geometry only
• .usd / .usda / .usdc — Universal Scene Description

Unsupported

.blend (Blender), .hip (Houdini), .ma/.mb (Maya) cannot be loaded directly. Export them to GLB/FBX/USD from the source tool first.

PAL imports static and animated FBX, GLB, and OBJ assets, including cameras with full animation tracks — position, rotation, focal length, and clip planes. Different DCCs need slightly different export settings to produce a clean FBX. Following the right checklist saves headaches like "camera is underground" or "rotation drifts mid-shot."

Blender

Default Blender camera FBX exports are mostly correct, but two settings need attention for animated cameras to import cleanly. Both are toggles that aren't on by default.

Step 1 — Bake constraints into explicit keyframes

If your camera uses a Track-To, Damped Track, or Look-At constraint to follow a target, those constraint-driven rotations don't automatically become F-curve keyframes that the FBX exporter can see. Without baking, the lens direction is lost wherever the constraint was the only thing aiming the camera.

1. Select the camera in the 3D viewport
2. Object → Animation → Bake Action…
3. In the popup:
   • Start Frame: 1 (or your shot's first frame)
   • End Frame: your shot's last frame
   • Frame Step: 1 (one keyframe per frame)
   • Only Selected: ON
   • Visual Keying: ON — bakes the constraint output into keyframes
   • Clear Constraints: ON
   • Clean Curves: OFF — important. Leaving this ON drops frames Blender considers redundant, causing playback gaps in PAL.
   • Channels: Location + Rotation
4. Click OK.

Verify in the Graph Editor: select the camera, open the Graph Editor, expand its rotation channels. You should see continuous dotted lines (one dot per frame) across the full range — no gaps.

Step 2 — FBX export with the right toggles

1. File → Export → FBX (.fbx)
2. In the export dialog:
   • Forward: -Z Forward
   • Up: Y Up
   • Use Space Transform: ON — required for animated cameras. Without it, the camera position can land "underground" in PAL.
   • Apply Unit: ON
   • Apply Scalings: All Local (or "FBX Units Scale" — either works)
3. Expand the Animation subpanel:
   • Bake Animation: ON
   • Force Start/End Keying: ON
   • Sampling Rate: 1.00
   • Simplify: 0.0 — important. The default of 1.0 silently drops keyframes during export, which can make animation diverge from your Blender preview.
4. Click Export FBX. Drop the file into PAL's viewport or Drive folder.

Common issues

• Camera underground in PAL. — Use Space Transform was off in FBX export. Re-export with it on.
• Position correct but framing diverges from Blender mid-shot. — A keyframe was dropped somewhere. Re-bake with Clean Curves OFF AND re-export with Simplify = 0. Both matter.
• Camera renders nothing but the HDRI background. — Camera far-clip is too tight. Blender camera Properties → Lens → Clip End.

Cube + camera in one FBX

You can export geometry and camera together. PAL splits them automatically: the geometry becomes a regular scene object; the camera becomes a camera proxy (cyan icon in the object list). Activate it via the VIEW dropdown in the camera bar to look through it. The two are independent objects in PAL after import — you can move, scale, or delete the geometry without affecting the camera, and vice versa.

Multiple cameras in one FBX all show up in the VIEW dropdown. While you're looking through one, the others appear as cyan wireframe markers in your scene so you can see where they sit — the markers are automatically hidden during render captures so they never appear in your beauty / depth / normal / alpha / matte passes.

Other DCCs

Coming soon — Maya, Cinema 4D, Houdini, 3ds Max, and Unreal export checklists. The pattern is similar: bake constraint-driven rotation into explicit keyframes, export FBX with the right axis convention, and disable any "simplify" / "redundancy" options that drop keyframes during export.

The AI Render panel takes the current viewport framing — your beauty pass and optional depth pass — and generates an AI image that respects the camera composition you authored in PAL. It's the bridge between layout work and final-look rendering.

Opening the panel

Click AI Render in the top toolbar. A floating, draggable panel appears at the top of the viewport. The active shot code shows next to the title.

Platform / Vendor / Model

Three cascading dropdowns. Selecting a Platform filters Vendor; selecting a Vendor filters Model. The same model is often available on multiple platforms — each has its own API key and cost. A live per-image cost readout next to the Model dropdown updates as you switch models so you know what each click will cost before you press RENDER.

Different models use different combinations of the available passes — depth-aware models consume the depth pass for spatial guidance, others use the beauty pass as a visual reference.

Prompt

Free-form text describing the look and style. Clicking Reset to shot description seeds it from the current shot's breakdown. Your prompt edits persist per-shot in local storage — closing and reopening the panel won't lose changes.

Live prompt preview

Right below the prompt editor, a preview panel shows what will reach the vendor — your typed prompt plus PAL’s auto-additions, colour-coded so you can see at a glance which part is yours and which the system is adding. Hover any coloured span for a tooltip; toggle a helper off and its span disappears in real time.

Prompt Helpers — editable, universal, optional

A collapsible section beneath the live preview gives you control over the auto-additions. Edit the text, reset to defaults, or disable a helper entirely. Both helpers persist edits locally per browser, and the panel remembers its collapsed / expanded state between sessions.

Sources tab — multi-reference images

For models that accept multiple reference images, a Sources tab next to the main prompt area lets you pick auxiliary images from your Drive library. The beauty pass from the viewport always occupies slot 1 (the structural base); selected sources fill the remaining slots up to the model's cap. Models without multi-ref support hide the tab.

Pass toggles (guidance only)

The checkboxes now only control which passes get sent to the AI model as guidance. All four passes (beauty, depth, normal, ID matte) are always archived to Drive and the render log regardless of the toggles, so the gallery viewer can always show them as pass tabs later.

• Depth — when checked, sent to the vendor as a structural guide. Required for ControlNet models on fal.ai; otherwise optional.
• Normal — currently archive-only.
• ID matte — per-object solid-colour pass (each PAL object gets a deterministic unique hue). Useful for per-object masks, downstream compositing, or feeding a segmentation model. Available as a first-class output in the Render & Export dialog and via the P-key live preview.
• Camera info position — off / before / after selector next to the pass toggles. When on (default after), PAL injects your current camera body, focal length, aperture and aspect ratio into the prompt so you don't have to repeat it. After is usually best — subject-first phrasing tends to produce stronger subject fidelity. Switch to off when working with edit-preserving models where extra trailing text can fight the composition.

All four passes land in the same Drive subfolder as source_beauty.png, source_depth.png, source_normal.png, source_matte.png.

Capture to (Render vs Sensor aspect)

Chooses which HUD frame the beauty pass is captured at:

• Render aspect (default, efficient) — capture at PAL's authoring resolution, matching the cyan dashed frame. No wasted vendor pixels; the AI output fills the render rectangle directly.
• Sensor aspect (cinema-style) — capture at the full sensor dimensions shown by the amber frame, send that to the vendor, then reformat to render aspect on return. Mirrors how a real camera records (sensor area) then delivers a crop (render area). Trade-off: the vendor burns resolution on the margins you'll crop away. Sensor long edge is capped at 4096 so uploads stay reasonable.

Reformat to PAL frame

Vendor output rarely matches PAL's exact render resolution or aspect. Choose how to reconcile:

• Crop — fill the PAL frame, losing edges from the vendor image (center crop). Default.
• Fit — letterbox with black bars so the entire vendor image is preserved.
• None — keep the vendor output untouched.

Output is upscaled (LANCZOS) to PAL resolution when smaller.

Fidelity (universal)

Picks which of the three Fidelity helper phrases gets prepended to your prompt. Applies to every platform. Edit the exact wording in the Prompt Helpers panel above.

• Loose — reference is inspiration only.
• Balanced — default. Match composition closely.
• Strict — preserve camera composition exactly, style/material changes only.

If you turn the Fidelity helper off in the Prompt Helpers panel, no phrase is injected regardless of the dropdown setting.

Enhance Prompt (all platforms)

Optional dropdown (Off / Gemini / Claude). When set, PAL rewrites your prompt through the chosen LLM before handing it to the image vendor — same behaviour as the Daily and PAL sheet tabs. The rewrite adds cinematic language and visual detail while preserving the specifics you wrote. Uses the API keys you've set in Settings.

Both the original and enhanced prompts are recorded on the render log and noted on the Log tab row so you can diff them later.

Guidance, Steps, Seed (model-dependent)

For models that expose them, three extra controls appear:

• Guidance — how strictly to follow the prompt. Higher = literal, lower = creative.
• Steps — refinement iterations. More = cleaner detail, slower.
• Seed — a fixed integer reproduces the same image for the same prompt + params. Blank = random each run.

Render button

Captures the configured passes from the current viewport at PAL render resolution, sends to the chosen model, and appends the result to the gallery. Status line shows progress; errors surface there too. A non-dismissable busy modal appears during the round-trip with a live elapsed-time counter so you always know how long the round-trip has been running (useful for identifying stuck requests vs. naturally slow ones).

Gallery

Thumbnails of previous renders appear below the controls. Toggle Current Shot vs All Shots scope. Renders are newest-first, numbered with a purple #1 badge so the latest is obvious. Click a thumbnail to open it in the floating viewer (below). Click the red × on a thumbnail to remove a single render. Use Clear All (red pill at the top) to wipe every thumbnail in the current scope — a styled confirm asks first, then DELETEs run in parallel. Drive files are never touched by either action; only the gallery reference is deleted.

Gallery viewer — pass switcher, resolution, render time

Clicking a thumbnail opens a floating, draggable, resizable viewer that replaces the old fullscreen modal. Position and size persist.

• Pass tabs under the title bar let you flip between the passes stored with that render: AI (the vendor output), Beauty, Depth, Normal, ID Matte. Only passes that exist on the record show up as tabs. Click or press 1–5 to switch.
• Resolution shows live in the title bar (1920×1080) from the image's natural pixel dimensions — useful for verifying the vendor matched your PAL render frame.
• Render time appears next to the model name in the meta row (e.g. flux-2-pro · 7.2s). Measured server-side from route entry to the render log write.
• Set as Starting Frame (amber button in the title bar) writes the currently-viewed pass to the PAL tab as the shot's Starting Frame column. Auto-exports the shot to the PAL tab if no row exists yet. See Shot Persistence and Starting Frame column for the downstream effect.
• Open in Drive — cyan link for the raw file.
• Arrow keys / on-screen ‹ › navigate between renders. Esc closes.

Persisted state

• Prompt, per shot
• Platform / Vendor / Model / Fit / Fidelity / Camera-info position / Depth toggles, globally
• Prompt Helpers: per-helper enable/disable, Camera suffix text, each Fidelity phrase (Loose / Balanced / Strict), and the collapsed/expanded state of the helpers panel
• Gallery viewer window position + size

Everything restores when the panel reopens, so your iteration state never resets.

Output files & data

Every successful render writes three places:

• Drive — the vendor output plus the source passes used to generate it land in a per-render subfolder under your project's render output folder, dated and named for the shot.
• Render gallery — the new render appears in the gallery panel for browsing later.
• Sheet → Log tab — one row per successful render with timestamp, status, generation mode, image tool, image URL, prompt, and notes.

Keys

API keys are stored encrypted per tenant. Set them via Settings → API Keys for the platforms you intend to use (Google, fal.ai, APIFrame, Dreamina).

Typical workflow

1. Block out the shot in PAL — camera, lens, subject placement.
2. Set the render resolution and aspect you want in Settings (or use the defaults).
3. Open AI Render. Pick Platform → Vendor → Model.
4. Describe the look in the prompt. Keep composition notes out — the beauty reference handles those.
5. Pick Fit / Crop and Fidelity. Leave depth on for ControlNet models.
6. Click RENDER. Wait 10–60s depending on vendor.
7. Iterate: tweak prompt, adjust fidelity, try a different model. The gallery keeps every version so you can compare.
8. Pick the winner and download from Drive, or reference the Log tab row in downstream pipeline runs.

PAL keeps a running local copy of every shot in your browser, alongside the normal cloud save. Every edit (object moved, keyframe added, lens changed, camera scrubbed) is captured immediately, even offline.

What's saved

• Scene graph — all PAL proxy objects with position, rotation (quaternion + Euler), scale, userData.
• Camera state — shot camera position, rotation, FOV, focal length.
• Camera system — active camera body, sensor mode, lens make/series/focal length, aperture, focus distance. Stored per shot — each shot remembers its own lens.
• Keyframes — all timeline keyframes with interpolation modes and quaternion data.
• Previz thumbnail — 320×180 JPEG captured on save for the EDIT reel.
• Extraction — the parsed shot description and derived metadata.

Recovery prompt

When you open a shot, PAL compares your local copy against the saved version. If the local copy has newer unsaved edits, a recovery modal appears:

Press Enter to Restore (default), Esc to Discard.

Per-shot lens

The camera system (camera body, sensor mode, lens focal length, aperture, focus distance) is stored on each shot individually. Switching shots applies that shot's saved lens.

Render passes match the HUD frame exactly

Every render pass captures exactly what's inside the cyan HUD frame at the configured render resolution — no extra margin, no scaling artefacts.

Starting Frame & End Frame columns (PAL tab)

The PAL tab has two explicit columns for video-restyle workflows:

The AI Render gallery viewer's Set as Starting Frame button writes to the Starting Frame column directly. Auto-exports the shot to the PAL tab if no row exists yet.

Legacy note: sheets that use the older Source Image header for this column still work without migration.

Settings auto-save on change.

Individual effects toggle independently: SSAO, Bloom, Lens Flare, Grid. All settings persist across sessions.

• Procedural — PAL generates a sky from your time-of-day and location settings. Cloud density and haze adjustable. No external file.
• HDRI — a high-dynamic-range image drives both the visible background and the scene lighting.
• Hybrid — HDRI lights the scene while the procedural sky fills the visible background.

Loading HDRIs

Mount a folder of .hdr or .exr files via Project Settings. Adjust Rotation to align the sun, Exposure to brighten/darken, Background Blur for DOF effect.

Frame the shot in Camera mode, click Capture Frame (or press F12) to save the view as a storyboard panel. Panels appear in the storyboard strip at the bottom; you can reorder, annotate, or delete them.

Click Export Storyboard to attach the panels to the corresponding shot records in LCBE, where they become part of the bid package.

PAL renders up to five passes per shot. Toggle each on or off in the Render & Export dialog:

• Beauty — full scene with sky/HDRI lighting
• Depth — distance gradient driven by the active camera's clip planes; gamma slider for midtone roll
• Normal — surface-normal pass
• Alpha — white objects on black background
• ID Matte — each scene object as a deterministic unique solid colour. Off by default; tick it when you need per-object masks.

Selected passes share a single Drive subfolder per render. URLs are written to the PAL sheet columns (Image Color / Depth / Normal / Alpha URL).

Use the P-key live preview (see section 5) to verify each pass before you commit to a full render.

Batch render

Check multiple shots in the render dialog — PAL switches shots automatically and renders each in sequence.

Server-side render

Click the green SERVER button in the render dialog to queue a headless render. You can close the browser tab and the render continues. A Discord notification fires on completion.

USD (Universal Scene Description) takes your PAL scene — cameras, animation, and object placement — directly into Houdini, Maya, Blender, or Unreal.

• USD — binary, smallest file, all USD-compatible tools
• USDA — human-readable text format
• USDZ — packaged format (Apple / AR)

Included in the USD file

• Camera with position, rotation, focal length, sensor size, keyframes
• All proxy geometry with correct position, rotation, scale
• Sun direction and intensity
• Proxy type labels and colour hints as custom properties
• Y-up or Z-up axis convention per your setting

Opening in DCC tools

• Blender — File → Import → USD
• Maya — File → Import, select USD format
• Houdini — USD Import SOP or stage-based workflow
• Unreal Engine — USD Stage Actor or content browser import

The camera appears as a proper camera object (not a mesh) with the correct focal length. Proxy objects come through as cubes with transform data — swap them for production assets in the DCC.