diff --git a/integrations/lingbot/lingbot/webrtc/web/request_session.js b/integrations/lingbot/lingbot/webrtc/web/request_session.js
index 4178a73d..846839f2 100644
--- a/integrations/lingbot/lingbot/webrtc/web/request_session.js
+++ b/integrations/lingbot/lingbot/webrtc/web/request_session.js
@@ -22,6 +22,9 @@ const remoteVideo = document.getElementById("remoteVideo")
const mockCanvas = document.getElementById("mockCanvas")
const firstFramePreview = document.getElementById("firstFramePreview")
const sceneCard = document.getElementById("sceneCard")
+const presetPicker = document.getElementById("presetPicker")
+const presetList = document.getElementById("presetList")
+const presetStatus = document.getElementById("presetStatus")
const firstFrameSourceRow = document.getElementById("firstFrameSourceRow")
const uploadModeButton = document.getElementById("uploadModeButton")
const urlModeButton = document.getElementById("urlModeButton")
@@ -67,6 +70,7 @@ let mockGenerationStarted = false
let mockChunkTimer = null
let actionStarted = false
let initialSceneLocked = false
+let presetSelectionInFlight = false
let promptEdited = false
let textEventsEdited = false
let firstFrameUrlEdited = false
@@ -193,6 +197,9 @@ function setInitialSceneLocked(locked) {
firstFrameUrlUpdateButton.disabled = locked
promptInput.disabled = locked
addTextEventButton.disabled = locked
+ for (const button of presetList.querySelectorAll("button")) {
+ button.disabled = locked || presetSelectionInFlight
+ }
for (const input of textEventList.querySelectorAll("input, textarea, button")) {
input.disabled = locked
}
@@ -209,6 +216,14 @@ function setFirstFrameInputMode(mode) {
}
function defaultFirstFrameName() {
+ const activePresetId = initialScene && initialScene.active_preset_id
+ const presets = Array.isArray(initialScene && initialScene.presets)
+ ? initialScene.presets
+ : []
+ const activePreset = presets.find((preset) => preset.preset_id === activePresetId)
+ if (activePreset && activePreset.label) {
+ return String(activePreset.label)
+ }
return initialScene && initialScene.has_first_frame ? "Example Image" : "Choose Image"
}
@@ -256,6 +271,91 @@ function refreshedPreviewUrl(url) {
return `${url}${separator}t=${firstFramePreviewRefreshToken}`
}
+function setPresetStatus(message = "", state = "idle") {
+ presetStatus.textContent = message
+ presetStatus.dataset.state = state
+}
+
+function renderPresetPicker() {
+ const presets = Array.isArray(initialScene && initialScene.presets)
+ ? initialScene.presets
+ : []
+ presetPicker.hidden = presets.length === 0
+ presetList.replaceChildren()
+ for (const preset of presets) {
+ const presetId = String(preset.preset_id || "").trim()
+ if (!presetId) {
+ continue
+ }
+ const label = String(preset.label || presetId)
+ const button = document.createElement("button")
+ button.className = "presetButton"
+ button.type = "button"
+ button.dataset.presetId = presetId
+ button.disabled = initialSceneLocked || presetSelectionInFlight
+ button.classList.toggle("is-active", initialScene.active_preset_id === presetId)
+ button.setAttribute("aria-pressed", initialScene.active_preset_id === presetId ? "true" : "false")
+ button.setAttribute("aria-label", `Use ${label} preset`)
+
+ const thumbnail = document.createElement("img")
+ thumbnail.className = "presetThumbnail"
+ thumbnail.src = String(preset.first_frame_url || "")
+ thumbnail.alt = ""
+ thumbnail.loading = "lazy"
+
+ const caption = document.createElement("span")
+ caption.className = "presetLabel"
+ caption.textContent = label
+ button.append(thumbnail, caption)
+ button.addEventListener("click", () => {
+ void selectPreset(presetId, label)
+ })
+ presetList.append(button)
+ }
+}
+
+async function selectPreset(presetId, label) {
+ if (initialSceneLocked || presetSelectionInFlight || mockMode) {
+ return
+ }
+ presetSelectionInFlight = true
+ setPresetStatus("Loading...", "pending")
+ renderPresetPicker()
+
+ try {
+ const form = new FormData()
+ form.append("preset_id", presetId)
+ const response = await fetch("/api/session/input", {
+ method: "POST",
+ body: form,
+ })
+ if (!response.ok) {
+ const text = (await response.text()).trim().replace(/^\d+:\s*/, "")
+ throw new Error(text || `preset selection failed (${response.status})`)
+ }
+
+ clearSelectedFirstFrameFile()
+ clearFirstFrameUrlInput()
+ setFirstFrameInputMode("url")
+ firstFrameSelectionCommitted = false
+ promptEdited = false
+ textEventsEdited = false
+ firstFrameUrlEdited = false
+ applyInitialScene(await response.json())
+ setPresetStatus("Selected", "success")
+ logEvent(`selected preset: ${label}`, { source: "client" })
+ } catch (error) {
+ setPresetStatus(error.message, "error")
+ logEvent(`preset selection failed: ${error.message}`, {
+ source: "client",
+ level: "error",
+ })
+ } finally {
+ presetSelectionInFlight = false
+ renderPresetPicker()
+ }
+}
+
function updateReadyPreview() {
const canPreview = !document.body.classList.contains("has-video")
const hasSelectedImage = selectedFirstFrameUrl !== null && firstFrameSelectionCommitted
@@ -283,7 +383,11 @@ function applyInitialScene(scene) {
}
const sceneImageUrl = typeof scene.image_url === "string"
? scene.image_url
- : (typeof scene.default_image_url === "string" ? scene.default_image_url : "")
+ : (
+ scene.input_source === "default" && typeof scene.default_image_url === "string"
+ ? scene.default_image_url
+ : ""
+ )
if (!selectedFirstFrameFile && !firstFrameUrlEdited && sceneImageUrl) {
firstFrameUrlInput.value = sceneImageUrl
setFirstFrameInputMode("url")
@@ -304,6 +408,7 @@ function applyInitialScene(scene) {
}
}
activeEventId = scene.active_event_id || null
+ renderPresetPicker()
if (!textEventsEdited) {
setTextEventDraftsFromCatalog(scene.event_catalog)
}
@@ -1001,6 +1106,13 @@ function disconnectSession({ notify = true } = {}) {
stopStatsPolling()
connected = false
actionStarted = false
+ inferenceInFlight = false
+ mockGenerationStarted = false
+ releaseAllKeys()
+ setInitialSceneLocked(false)
+ setStatus("Idle", "idle")
+ setFlow("waiting")
+ connectButton.textContent = mockMode ? "Start Mock Session" : "Connect Session"
updateReadyPreview()
connectButton.disabled = false
if (notify && controlChannel && controlChannel.readyState === "open") {
@@ -1016,6 +1128,8 @@ function disconnectSession({ notify = true } = {}) {
if (peerConnection) {
peerConnection.close()
}
+ remoteVideo.srcObject = null
+ setVideoVisible(false)
}
async function connectSession() {
@@ -1067,6 +1181,8 @@ async function connectSession() {
logEvent(`connection_state=${state}`, { source: "client" })
if (state === "connected") {
connected = true
+ connectButton.disabled = false
+ connectButton.textContent = "Disconnect Session"
setStatus("Waiting", "waiting")
setFlow("connected; waiting for input")
startStatsPolling()
@@ -1079,8 +1195,10 @@ async function connectSession() {
if (["failed", "closed", "disconnected"].includes(state)) {
connected = false
actionStarted = false
+ setInitialSceneLocked(false)
updateReadyPreview()
connectButton.disabled = false
+ connectButton.textContent = "Connect Session"
stopHeartbeat()
stopStatsPolling()
setStatus(state === "failed" ? "Error" : "Idle", state === "failed" ? "error" : "idle")
@@ -1369,12 +1487,15 @@ async function startMockSession() {
setStatus("Connecting", "connecting")
setFlow("mock warmup")
logEvent("connecting to mock server...", { source: "client" })
+ disconnecting = false
actionStarted = false
await uploadSessionInputIfNeeded()
await new Promise((resolve) => {
window.setTimeout(resolve, 260)
})
connected = true
+ connectButton.disabled = false
+ connectButton.textContent = "Disconnect Session"
metrics.targetFps = 16
metrics.resolution = "1280x720"
metrics.model = "lingbot-world-v2-14b-causal-fast-taehv-window15-sink3"
@@ -1516,7 +1637,11 @@ function initialize() {
}
connectButton.addEventListener("click", () => {
- void connectSession()
+ if (connected) {
+ disconnectSession()
+ } else {
+ void connectSession()
+ }
})
clearEventButton.addEventListener("click", () => {
sendTextEvent(activeEventId || "clear", "clear")
diff --git a/integrations/lingbot/pyproject.toml b/integrations/lingbot/pyproject.toml
index f9e39cd9..6bdea26d 100644
--- a/integrations/lingbot/pyproject.toml
+++ b/integrations/lingbot/pyproject.toml
@@ -62,6 +62,11 @@ exclude = ["tests"]
"*.js",
"assets/*.svg",
]
+"lingbot.webrtc" = [
+ "presets/*/first_frame.png",
+ "presets/*/prompt.txt",
+ "presets/*/event_texts.json",
+]
[tool.uv]
managed = true
diff --git a/integrations/lingbot/scripts/launch_webrtc.sh b/integrations/lingbot/scripts/launch_webrtc.sh
index 391919c7..263be75b 100755
--- a/integrations/lingbot/scripts/launch_webrtc.sh
+++ b/integrations/lingbot/scripts/launch_webrtc.sh
@@ -13,6 +13,12 @@ FPS="${FPS:-16}"
EXAMPLE_IDX="${EXAMPLE_IDX:-0}"
VIDEO_HEIGHT="${VIDEO_HEIGHT:-352}"
VIDEO_WIDTH="${VIDEO_WIDTH:-640}"
+PRESET_ASSETS_DIR="${PRESET_ASSETS_DIR:-}"
+
+PRESET_ARGS=()
+if [[ -n "${PRESET_ASSETS_DIR}" ]]; then
+ PRESET_ARGS+=(--preset-assets-dir "${PRESET_ASSETS_DIR}")
+fi
cd "$(dirname "${BASH_SOURCE[0]}")/../../.."
@@ -25,4 +31,5 @@ exec uv run --no-sync python -m lingbot.webrtc.server \
--fps "${FPS}" \
--video-height "${VIDEO_HEIGHT}" \
--video-width "${VIDEO_WIDTH}" \
- --example-idx "${EXAMPLE_IDX}"
+ --example-idx "${EXAMPLE_IDX}" \
+ "${PRESET_ARGS[@]}"
diff --git a/integrations/lingbot/tests/test_preset_assets.py b/integrations/lingbot/tests/test_preset_assets.py
new file mode 100644
index 00000000..628f1110
--- /dev/null
+++ b/integrations/lingbot/tests/test_preset_assets.py
@@ -0,0 +1,187 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""CPU checks for LingBot WebRTC preset assets."""
+
+from __future__ import annotations
+
+import json
+import sys
+from argparse import Namespace
+from pathlib import Path
+
+import pytest
+from lingbot.webrtc import server
+
+pytestmark = pytest.mark.ci_cpu
+
+
+def _write_preset(directory: Path, *, wrapped_events: bool = False) -> None:
+ """Write a minimal valid preset-assets directory."""
+ directory.mkdir()
+ (directory / server.PRESET_FIRST_FRAME_FILENAME).write_bytes(b"\x89PNG\r\n\x1a\n")
+ (directory / server.PRESET_PROMPT_FILENAME).write_text(
+ "A quiet portrait by a sunny window.\n",
+ encoding="utf-8",
+ )
+ events: object = [
+ {
+ "event_id": "smile",
+ "label": "Smile",
+ "prompt": "A warm smile forms on her face.",
+ "category": "portrait",
+ }
+ ]
+ if wrapped_events:
+ events = {"events": events}
+ (directory / server.PRESET_TEXT_EVENTS_FILENAME).write_text(
+ json.dumps(events),
+ encoding="utf-8",
+ )
+
+
+def _server_args(preset_assets_dir: Path | None) -> Namespace:
+ """Build WebRTC CLI arguments without parsing process arguments."""
+ return Namespace(
+ config_name="lingbot-world-v2-14b-causal-fast",
+ no_compile=True,
+ device="cuda:0",
+ warmup_chunks=0,
+ warmup_timeout_s=600.0,
+ video_height=352,
+ video_width=640,
+ example_idx=0,
+ preset_assets_dir=preset_assets_dir,
+ )
+
+
+@pytest.mark.parametrize("wrapped_events", [False, True])
+def test_load_preset_assets_accepts_supported_event_layouts(
+ tmp_path: Path,
+ wrapped_events: bool,
+) -> None:
+ """Load direct and object-wrapped event catalogs."""
+ preset_dir = tmp_path / "preset"
+ _write_preset(preset_dir, wrapped_events=wrapped_events)
+
+ resolved_dir, events = server.load_preset_assets(preset_dir)
+
+ assert resolved_dir == preset_dir.resolve()
+ assert [event.event_id for event in events] == ["smile"]
+ assert events[0].prompt == "A warm smile forms on her face."
+
+
+def test_load_preset_assets_reports_missing_files(tmp_path: Path) -> None:
+ """List every required asset missing from an incomplete preset."""
+ preset_dir = tmp_path / "preset"
+ preset_dir.mkdir()
+
+ with pytest.raises(
+ ValueError, match="first_frame.png, prompt.txt, event_texts.json"
+ ):
+ server.load_preset_assets(preset_dir)
+
+
+def test_build_runtime_config_uses_preset_assets(tmp_path: Path) -> None:
+ """Override the default example frame and event catalog with a preset."""
+ preset_dir = tmp_path / "preset"
+ _write_preset(preset_dir)
+
+ config = server.build_runtime_config(_server_args(preset_dir))
+
+ assert config.example_data_dir == preset_dir.resolve()
+ assert config.first_frame_filename == "first_frame.png"
+ assert config.prompt_filename == "prompt.txt"
+ assert config.default_image_url is None
+ assert config.default_preset_id is None
+ assert [event.event_id for event in config.text_events] == ["smile"]
+
+
+def test_build_runtime_config_identifies_bundled_default_preset() -> None:
+ """Mark a bundled launch preset as selected in initial-scene metadata."""
+ preset_dir = (
+ Path(server.__file__).resolve().parent / "presets" / "golden-hour-portrait"
+ )
+
+ config = server.build_runtime_config(_server_args(preset_dir))
+
+ assert config.default_preset_id == "golden-hour-portrait"
+
+
+def test_parse_args_accepts_preset_assets_directory(
+ monkeypatch: pytest.MonkeyPatch,
+ tmp_path: Path,
+) -> None:
+ """Expose the preset directory through both CLI spellings."""
+ monkeypatch.setattr(
+ sys,
+ "argv",
+ ["lingbot.webrtc.server", "--preset-assets-dir", str(tmp_path)],
+ )
+
+ args = server.parse_args()
+
+ assert args.preset_assets_dir == tmp_path
+
+
+def test_bundled_preset_catalog_exposes_picker_metadata() -> None:
+ """Load all bundled presets in their stable UI order."""
+ presets = server.load_bundled_presets()
+
+ assert [preset.preset_id for preset in presets] == list(server.BUNDLED_PRESET_IDS)
+ assert all(preset.prompt for preset in presets)
+ assert all(preset.first_frame.data.startswith(b"\x89PNG") for preset in presets)
+ assert all(preset.text_events for preset in presets)
+ assert presets[0].as_public_dict() == {
+ "preset_id": "golden-hour-portrait",
+ "label": "Golden Hour Portrait",
+ "first_frame_url": "/api/presets/golden-hour-portrait/first_frame",
+ }
+
+
+@pytest.mark.parametrize(
+ ("preset_name", "expected_event_ids"),
+ [
+ (
+ "golden-hour-portrait",
+ ["hair-tuck", "subtle-smile", "head-turn"],
+ ),
+ (
+ "moonlit-portal",
+ ["portal-awakens", "fireflies-gather", "storm-approaches"],
+ ),
+ (
+ "cozy-reading-room",
+ ["fire-brightens", "pages-turn", "rain-intensifies"],
+ ),
+ (
+ "misty-dinosaur-valley",
+ ["dinosaur-raises-head", "flock-crosses-sky", "mist-rolls-in"],
+ ),
+ ],
+)
+def test_bundled_preset_is_valid(
+ preset_name: str,
+ expected_event_ids: list[str],
+) -> None:
+ """Keep every bundled preset synchronized with the asset schema."""
+ preset_dir = Path(server.__file__).resolve().parent / "presets" / preset_name
+
+ resolved_dir, events = server.load_preset_assets(preset_dir)
+
+ assert resolved_dir == preset_dir
+ assert (preset_dir / "first_frame.png").read_bytes().startswith(b"\x89PNG")
+ assert (preset_dir / "prompt.txt").read_text(encoding="utf-8").strip()
+ assert [event.event_id for event in events] == expected_event_ids
diff --git a/integrations/lingbot/tests/test_server_routes.py b/integrations/lingbot/tests/test_server_routes.py
index 23e65cad..38368fa9 100644
--- a/integrations/lingbot/tests/test_server_routes.py
+++ b/integrations/lingbot/tests/test_server_routes.py
@@ -79,7 +79,8 @@ def set_pending_session_input(self, session_input: LingbotSessionInput) -> None:
**self.initial_scene,
"prompt": session_input.prompt or self.initial_scene["prompt"],
"event_catalog": event_catalog,
- "input_source": "uploaded",
+ "input_source": "preset" if session_input.preset_id else "uploaded",
+ "active_preset_id": session_input.preset_id,
}
async def preload_runtime(self) -> None:
@@ -223,7 +224,11 @@ async def test_initial_scene_route_returns_preview_metadata() -> None:
response = await client.get("/api/session/initial_scene")
payload = await response.json()
assert response.status == 200
- assert payload == manager.initial_scene
+ assert payload["prompt"] == manager.initial_scene["prompt"]
+ assert payload["active_preset_id"] is None
+ assert [preset["preset_id"] for preset in payload["presets"]] == list(
+ lingbot_server.BUNDLED_PRESET_IDS
+ )
finally:
await client.close()
@@ -242,6 +247,68 @@ async def test_first_frame_route_serves_manager_image() -> None:
await client.close()
+@pytest.mark.asyncio
+async def test_bundled_preset_preview_route_serves_png() -> None:
+ manager = FakeSessionManager()
+ client = await _build_client(manager)
+ try:
+ response = await client.get("/api/presets/golden-hour-portrait/first_frame")
+ body = await response.read()
+
+ assert response.status == 200
+ assert response.headers["Content-Type"] == "image/png"
+ assert body.startswith(b"\x89PNG\r\n\x1a\n")
+ finally:
+ await client.close()
+
+
+@pytest.mark.asyncio
+async def test_session_input_selects_bundled_preset() -> None:
+ manager = FakeSessionManager()
+ client = await _build_client(manager)
+ try:
+ form = FormData()
+ form.add_field("preset_id", "moonlit-portal")
+
+ response = await client.post("/api/session/input", data=form)
+ payload = await response.json()
+
+ assert response.status == 200
+ assert payload["active_preset_id"] == "moonlit-portal"
+ assert payload["input_source"] == "preset"
+ assert len(manager.pending_inputs) == 1
+ session_input = manager.pending_inputs[0]
+ assert session_input.preset_id == "moonlit-portal"
+ assert session_input.prompt is not None and "portal" in session_input.prompt
+ assert session_input.first_frame_image_bytes is not None
+ assert session_input.first_frame_image_bytes.startswith(b"\x89PNG\r\n\x1a\n")
+ assert session_input.first_frame_content_type == "image/png"
+ assert session_input.text_events is not None
+ assert [event.event_id for event in session_input.text_events] == [
+ "portal-awakens",
+ "fireflies-gather",
+ "storm-approaches",
+ ]
+ finally:
+ await client.close()
+
+
+@pytest.mark.asyncio
+async def test_session_input_rejects_unknown_preset() -> None:
+ manager = FakeSessionManager()
+ client = await _build_client(manager)
+ try:
+ form = FormData()
+ form.add_field("preset_id", "missing-preset")
+
+ response = await client.post("/api/session/input", data=form)
+
+ assert response.status == 400
+ assert manager.pending_inputs == []
+ finally:
+ await client.close()
+
+
@pytest.mark.asyncio
async def test_blocking_session_routes_use_worker_thread(
monkeypatch: pytest.MonkeyPatch,
diff --git a/integrations/lingbot/tests/test_webrtc_runtime.py b/integrations/lingbot/tests/test_webrtc_runtime.py
index e3ac8f65..2245312a 100644
--- a/integrations/lingbot/tests/test_webrtc_runtime.py
+++ b/integrations/lingbot/tests/test_webrtc_runtime.py
@@ -231,6 +231,7 @@ def _load_default_prompt(self) -> str:
assert scene["capabilities"] == {"text_events": True}
assert scene["active_event_id"] is None
+ assert scene["active_preset_id"] is None
assert scene["event_catalog"] == [
event.as_public_dict() for event in session.DEFAULT_TEXT_EVENTS
]
@@ -291,6 +292,40 @@ def _load_default_prompt(self) -> str:
assert scene["event_catalog"] == [custom_events[0].as_public_dict()]
+def test_initial_scene_tracks_selected_preset(
+ monkeypatch: pytest.MonkeyPatch,
+) -> None:
+ """Advertise the default and pending preset selections."""
+
+ class _FakeRuntime:
+ _active_event_id = None
+
+ def __init__(self, config: LingbotRuntimeConfig) -> None:
+ self.config = config
+
+ def _load_default_prompt(self) -> str:
+ return "drive through a city"
+
+ monkeypatch.setattr(session, "LingbotInferenceRuntime", _FakeRuntime)
+ manager = LingbotWebRTCSessionManager(
+ runtime_config=LingbotRuntimeConfig(
+ device="cpu",
+ warmup_chunks=0,
+ default_preset_id="golden-hour-portrait",
+ )
+ )
+
+ assert manager.get_initial_scene()["active_preset_id"] == "golden-hour-portrait"
+
+ manager.set_pending_session_input(
+ session.LingbotSessionInput(preset_id="moonlit-portal")
+ )
+
+ scene = manager.get_initial_scene()
+ assert scene["active_preset_id"] == "moonlit-portal"
+ assert scene["input_source"] == "preset"
+
+
def test_pending_remote_first_frame_is_fetched_once_and_cached(
monkeypatch: pytest.MonkeyPatch,
) -> None:
@@ -400,6 +435,36 @@ def _fail_remote_fetch(image_url: str) -> object:
assert runtime._prompt == "follow a coastal highway"
+def test_prepare_session_input_state_accepts_empty_prompt(
+ monkeypatch: pytest.MonkeyPatch,
+) -> None:
+ """Encode an empty base prompt when preset ``prompt.txt`` is empty."""
+ runtime = session.LingbotInferenceRuntime(
+ config=LingbotRuntimeConfig(device="cpu", warmup_chunks=0)
+ )
+ runtime._device = torch.device("cpu")
+ encoded_texts: list[tuple[str, ...]] = []
+
+ monkeypatch.setattr(runtime, "_load_default_prompt", lambda: "")
+ monkeypatch.setattr(runtime, "_load_default_first_frame_rgb", object)
+ monkeypatch.setattr(
+ runtime,
+ "_first_frame_to_tensor",
+ lambda image_rgb: torch.zeros((1, 3, 2, 2)),
+ )
+
+ def _encode(texts: list[str]) -> torch.Tensor:
+ encoded_texts.append(tuple(texts))
+ return torch.zeros((len(texts), 1, 2))
+
+ monkeypatch.setattr(runtime, "_encode_text_embeddings_sync", _encode)
+
+ runtime._prepare_session_input_state(None)
+
+ assert runtime._prompt == ""
+ assert encoded_texts == [("",)]
+
+
@pytest.mark.asyncio
async def test_event_message_dispatches_to_runtime_and_acknowledges(
monkeypatch: pytest.MonkeyPatch,