Studio: video inference backend with LTX-2 and LTX-2.3 support

scripts/scan_packages_baseline.json

@@ -226,6 +226,22 @@
    "evidence": "FS: L528:         history: list[Response] | None = None, sha256:f56272dccd651b2644aa41ef6e688e211462427aad07fef5150240ec7347446e\nNetwork: L9: import urllib.request | L1243:     class _CookieCompatRequest(urllib.request.Request):",
    "evidence_hash": "b32f79e58c938680d89efa74113eeba76c9fc5aedf5de18086f93bef274c4bda"
   },
+  {
+   "package": "huggingface-hub",
+   "file": "huggingface_hub/_sandbox.py",
+   "check": "C2 polling/beaconing loop detected",
+   "severity": "CRITICAL",
+   "evidence": "L1179:             while True: sha256:33ceddf9e42aae207e891e97808c518e92a0b27ab60e4326256717bfb25a3a38",
+   "evidence_hash": "802fd41d8bb17bf425e99d128c0351c820103a5efb74690a4086e542a71437b8"
+  },
+  {
+   "package": "huggingface-hub",
+   "file": "huggingface_hub/_sandbox.py",
+   "check": "Writes to /tmp and executes (staged dropper)",
+   "severity": "CRITICAL",
+   "evidence": "L83: d=/tmp/.sbx-server\nL84: if command -v wget >/dev/null 2>&1; then wget -q --header \"Authorization: Bearer $SBX_DL_TOKEN\" -O \"$d\" \"$SBX_SERVER_URL\"\nL85: elif command -v curl >/dev/null 2>&1; then curl -fsSL -H \"Authorization: Bearer $SBX_DL_TOKEN\" -o \"$d\" \"$SBX_SERVER_URL\"\nL86: else cp \"$SBX_SERVER_MOUNT/sbx-server\" \"$d\"; fi\nL87: chmod +x \"$d\"",
+   "evidence_hash": "6908a3fe328fa94ee22a119998d6ad07cfa1ba4efa2628acf240f4204fd76e22"
+  },
   {
    "package": "huggingface-hub",
    "file": "huggingface_hub/hf_api.py",
@@ -242,6 +258,14 @@
    "evidence": "L4600:         while True: sha256:f4a851312a1832efe1b435aa1275a82184e19cc3f47e2cd244373d56c11de272",
    "evidence_hash": "dc8fcf44788e32f42d1cc2eb0e2deb55eb2dbf2c3a55909a7d503e450f45e602"
   },
+  {
+   "package": "huggingface-hub",
+   "file": "huggingface_hub/hf_api.py",
+   "check": "C2 polling/beaconing loop detected",
+   "severity": "CRITICAL",
+   "evidence": "L4613:         while True: sha256:f764b6ca3118b23c7c0e670e77178c022a6905f825d7df6e528545fa10aae8f6",
+   "evidence_hash": "9c85d50c227285fa8dc69512999cbb082258cda4b299c7d0e0f69f5aff7accd4"
+  },
   {
    "package": "huggingface-hub",
    "file": "huggingface_hub/hf_api.py",
@@ -266,6 +290,14 @@
    "evidence": "L298:     while True: sha256:6b8e5e569594caf7c4eca6137646dae471a7c3aae7294096cf876f30b5f90306",
    "evidence_hash": "c066cc27bce31ee7b6ce07411ee7a7d9ecfbf3aafc8848f6641fabfe522a7703"
   },
+  {
+   "package": "huggingface-hub",
+   "file": "huggingface_hub/utils/_http.py",
+   "check": "C2 polling/beaconing loop detected",
+   "severity": "CRITICAL",
+   "evidence": "L462:     while True: sha256:c75d1ee228cf7703a8c28551d649395a1f89f69a3aba69413f5bbcbd10c31958",
+   "evidence_hash": "d4d5f83fed39b87898cf776d5dad0bf1a6388a932f5fb7997d1070b50e46213e"
+  },
   {
    "package": "ipython",
    "file": "IPython/core/debugger.py",

studio/backend/core/inference/diffusion_memory.py

@@ -269,6 +269,26 @@ def estimate_image_runtime_mib(
     return max(1024, int(8192 * max(0.25, pixel_scale) * multiplier))
 
 
+def estimate_video_runtime_mib(
+    *, width: Optional[int], height: Optional[int], num_frames: Optional[int]
+) -> int:
+    """Per-call activation / latent / decode headroom for a video generation.
+
+    The image estimator is pixel-area only and badly undershoots video: latents
+    carry a frames dimension and the VAE DECODE is the peak -- the decoded clip
+    materialises as num_frames full-resolution fp32 frames plus the decoder's
+    intermediates, typically dwarfing the denoise activations. Scale by the
+    decoded-clip footprint (frames x H x W x 3 x 4 bytes) with a 3x factor for
+    decoder intermediates + the PIL/tensor copy held during export, on top of a
+    fixed denoise-side base.
+    """
+    w = max(64, int(width or 768))
+    h = max(64, int(height or 512))
+    frames = max(1, int(num_frames or 121))
+    decoded_mib = (frames * w * h * 3 * 4) / float(1024 * 1024)
+    return max(3072, int(4096 + 3.0 * decoded_mib))
+
+
 def _safe_device_budget_mib(memory: DeviceMemory) -> Optional[int]:
     """Free memory minus a headroom reserve, so a plan that "fits" leaves room for
     fragmentation and other tenants. None when free memory is unknown."""

studio/backend/core/inference/gpu_arbiter.py

@@ -24,6 +24,7 @@
 
 CHAT = "chat"
 DIFFUSION = "diffusion"
+VIDEO = "video"
 
 _lock = threading.Lock()
 _owner: Optional[str] = None
@@ -61,8 +62,16 @@ def _evict_diffusion() -> None:
     get_active_diffusion_engine().unload()
 
 
-# Patchable in tests via monkeypatch.setitem.
-_EVICTORS = {CHAT: _evict_chat, DIFFUSION: _evict_diffusion}
+def _evict_video() -> None:
+    from core.inference.video import get_video_backend
+    get_video_backend().unload()
+
+
+# Patchable in tests via monkeypatch.setitem. Ownership is exclusive -- only one
+# owner holds the GPU at a time -- so acquire_for's evict-the-current-owner
+# already generalises to any number of registered owners (chat / image / video
+# all evict whichever of the others currently holds the GPU).
+_EVICTORS = {CHAT: _evict_chat, DIFFUSION: _evict_diffusion, VIDEO: _evict_video}
 
 
 def acquire_for(owner: str) -> None: