Add LTX-2.3 overlay metadata

README.md

@@ -0,0 +1,14 @@
+---
+library_name: sglang-diffusion
+tags:
+- sglang-diffusion
+- overlay
+- ltx-2
+- ltx-2-3
+---
+
+# LTX-2.3 Overlay
+
+Overlay metadata repo for using `Lightricks/LTX-2.3` with SGLang Diffusion native LTX2 support.
+
+This repo contains only overlay metadata and materialization logic. Source weights remain in `Lightricks/LTX-2.3`.

_overlay/materialize.py

@@ -0,0 +1,71 @@
+import os
+
+from huggingface_hub import snapshot_download
+
+from sglang.multimodal_gen.runtime.utils.model_overlay import (
+    _copytree_link_or_copy,
+    _ensure_dir,
+    _link_or_copy_file,
+)
+
+AUXILIARY_MODEL_ID = "Lightricks/LTX-2"
+AUXILIARY_PATTERNS = [
+    "audio_vae/**",
+    "connectors/**",
+    "scheduler/**",
+    "text_encoder/**",
+    "tokenizer/**",
+    "transformer/config.json",
+    "vae/**",
+    "vocoder/**",
+]
+
+
+def materialize(
+    *,
+    overlay_dir: str,
+    source_dir: str,
+    output_dir: str,
+    manifest: dict,
+) -> None:
+    _ = overlay_dir, manifest
+
+    auxiliary_dir = snapshot_download(
+        repo_id=AUXILIARY_MODEL_ID,
+        allow_patterns=AUXILIARY_PATTERNS,
+        max_workers=8,
+    )
+
+    for component_name in (
+        "audio_vae",
+        "connectors",
+        "scheduler",
+        "text_encoder",
+        "tokenizer",
+        "vae",
+        "vocoder",
+    ):
+        _copytree_link_or_copy(
+            os.path.join(auxiliary_dir, component_name),
+            os.path.join(output_dir, component_name),
+        )
+
+    transformer_dir = os.path.join(output_dir, "transformer")
+    _ensure_dir(transformer_dir)
+    _link_or_copy_file(
+        os.path.join(auxiliary_dir, "transformer", "config.json"),
+        os.path.join(transformer_dir, "config.json"),
+    )
+    _link_or_copy_file(
+        os.path.join(source_dir, "ltx-2.3-22b-dev.safetensors"),
+        os.path.join(transformer_dir, "ltx-2.3-22b-dev.safetensors"),
+    )
+
+    _link_or_copy_file(
+        os.path.join(source_dir, "ltx-2.3-22b-distilled-lora-384.safetensors"),
+        os.path.join(output_dir, "ltx-2.3-22b-distilled-lora-384.safetensors"),
+    )
+    _link_or_copy_file(
+        os.path.join(source_dir, "ltx-2.3-spatial-upscaler-x2-1.1.safetensors"),
+        os.path.join(output_dir, "ltx-2.3-spatial-upscaler-x2-1.1.safetensors"),
+    )

_overlay/overlay_manifest.json

@@ -0,0 +1,15 @@
+{
+  "source_model_id": "Lightricks/LTX-2.3",
+  "source_allow_patterns": [
+    "ltx-2.3-22b-dev.safetensors",
+    "ltx-2.3-22b-distilled-lora-384.safetensors",
+    "ltx-2.3-spatial-upscaler-x2-1.1.safetensors"
+  ],
+  "required_source_files": [
+    "ltx-2.3-22b-dev.safetensors",
+    "ltx-2.3-22b-distilled-lora-384.safetensors",
+    "ltx-2.3-spatial-upscaler-x2-1.1.safetensors"
+  ],
+  "custom_materializer": "_overlay/materialize.py",
+  "materializer_version": "v1"
+}

model_index.json

@@ -0,0 +1,36 @@
+{
+  "_class_name": "LTX2Pipeline",
+  "_diffusers_version": "0.37.0.dev0",
+  "audio_vae": [
+    "diffusers",
+    "AutoencoderKLLTX2Audio"
+  ],
+  "connectors": [
+    "ltx2",
+    "LTX2TextConnectors"
+  ],
+  "scheduler": [
+    "diffusers",
+    "FlowMatchEulerDiscreteScheduler"
+  ],
+  "text_encoder": [
+    "transformers",
+    "Gemma3ForConditionalGeneration"
+  ],
+  "tokenizer": [
+    "transformers",
+    "GemmaTokenizerFast"
+  ],
+  "transformer": [
+    "diffusers",
+    "LTX2VideoTransformer3DModel"
+  ],
+  "vae": [
+    "diffusers",
+    "AutoencoderKLLTX2Video"
+  ],
+  "vocoder": [
+    "ltx2",
+    "LTX2Vocoder"
+  ]
+}