Require HF login and user gated-model access

.playwright-cli/console-2026-05-20T20-23-31-945Z.log

@@ -0,0 +1,3 @@
+[    1437ms] [ERROR] Failed to load resource: the server responded with a status of 400 () @ https://huggingface.co/api/spaces/by-subdomain/owenisas-stable-audio-3-lab:0
+[    1700ms] [WARNING] Failed to execute 'postMessage' on 'DOMWindow': The target origin provided ('https://huggingface.co') does not match the recipient window's origin ('https://owenisas-stable-audio-3-lab.hf.space'). @ https://owenisas-stable-audio-3-lab.hf.space/assets/Index-Be9xvQ3a.js:1
+[    1740ms] [ERROR] Failed to load resource: the server responded with a status of 404 () @ https://huggingface.co/api/organizations/owenisas/avatar:0

.playwright-cli/page-2026-05-20T20-23-33-278Z.yml

@@ -0,0 +1,4 @@
+- main [ref=e4]:
+  - generic [ref=e5]:
+    - img [ref=e9]
+    - paragraph [ref=e20]: Loading...

.playwright-cli/page-2026-05-20T20-23-51-604Z.yml

@@ -0,0 +1,224 @@
+- generic [ref=e1]:
+  - main [ref=e4]:
+    - generic [ref=e21]:
+      - generic [ref=e23]:
+        - heading "Stable Audio 3 Lab" [level=1] [ref=e28]
+        - generic [ref=e29]:
+          - generic [ref=e30]:
+            - generic [ref=e31]:
+              - button [ref=e32] [cursor=pointer]: Generate
+              - button [ref=e33] [cursor=pointer]: Autoencoder
+              - button [ref=e34] [cursor=pointer]: Coverage
+            - tablist [ref=e35]:
+              - tab "Generate" [selected] [ref=e36] [cursor=pointer]
+              - tab "Autoencoder" [ref=e37] [cursor=pointer]
+              - tab "Coverage" [ref=e38] [cursor=pointer]
+          - tabpanel [ref=e39]:
+            - generic [ref=e41]:
+              - generic [ref=e42]:
+                - generic [ref=e43]:
+                  - generic [ref=e45]:
+                    - generic [ref=e46]: Model
+                    - generic [ref=e49]:
+                      - listbox "Model" [ref=e50]: Stable Audio 3 Small SFX
+                      - generic:
+                        - img
+                  - generic [ref=e52]:
+                    - generic [ref=e53]: Prompt
+                    - textbox "Prompt" [ref=e55]:
+                      - /placeholder: ""
+                      - text: Close binaural rain on a window, soft cloth movement, detailed texture
+                  - generic [ref=e57]:
+                    - generic [ref=e58]: Negative prompt
+                    - textbox "Negative prompt" [ref=e60]:
+                      - /placeholder: ""
+                - generic [ref=e62]:
+                  - generic [ref=e64]:
+                    - generic [ref=e65]:
+                      - generic [ref=e67]: Duration
+                      - generic [ref=e68]:
+                        - spinbutton "number input for Duration" [ref=e69]: "8"
+                        - button "Reset to default value" [ref=e70] [cursor=pointer]: ↺
+                    - generic [ref=e71]:
+                      - generic [ref=e72]: "1"
+                      - slider "range slider for Duration" [ref=e73] [cursor=pointer]: "8"
+                      - generic [ref=e74]: "120"
+                  - generic [ref=e76]:
+                    - generic [ref=e77]:
+                      - generic [ref=e79]: Steps
+                      - generic [ref=e80]:
+                        - spinbutton "number input for Steps" [ref=e81]: "8"
+                        - button "Reset to default value" [ref=e82] [cursor=pointer]: ↺
+                    - generic [ref=e83]:
+                      - generic [ref=e84]: "1"
+                      - slider "range slider for Steps" [ref=e85] [cursor=pointer]: "8"
+                      - generic [ref=e86]: "100"
+                  - generic [ref=e88]:
+                    - generic [ref=e89]:
+                      - generic [ref=e91]: CFG
+                      - generic [ref=e92]:
+                        - spinbutton "number input for CFG" [ref=e93]: "1"
+                        - button "Reset to default value" [ref=e94] [cursor=pointer]: ↺
+                    - generic [ref=e95]:
+                      - generic [ref=e96]: "0"
+                      - slider "range slider for CFG" [ref=e97] [cursor=pointer]: "1"
+                      - generic [ref=e98]: "12"
+                - generic [ref=e100]:
+                  - generic [ref=e102]:
+                    - generic [ref=e103]: Sampler
+                    - generic [ref=e106]:
+                      - listbox "Sampler" [ref=e107]: pingpong
+                      - generic:
+                        - img
+                  - generic [ref=e109]:
+                    - generic [ref=e110]: Seed
+                    - spinbutton "Seed" [ref=e111]: "-1"
+                - generic [ref=e113]:
+                  - generic [ref=e115] [cursor=pointer]:
+                    - checkbox "Chunked decode" [checked] [ref=e116]
+                    - generic [ref=e117]: Chunked decode
+                  - generic [ref=e119] [cursor=pointer]:
+                    - checkbox "CPU override" [ref=e120]
+                    - generic [ref=e121]: CPU override
+                - generic [ref=e122]:
+                  - button "Generate" [ref=e123] [cursor=pointer]
+                  - button "Unload" [ref=e124] [cursor=pointer]
+                  - button "Runtime" [active] [ref=e125] [cursor=pointer]
+              - generic [ref=e126]:
+                - generic [ref=e127]:
+                  - generic [ref=e128]:
+                    - generic:
+                      - generic:
+                        - img
+                      - text: Model info
+                  - button "Copy" [ref=e130] [cursor=pointer]:
+                    - img [ref=e132]
+                  - generic [ref=e136]:
+                    - generic [ref=e137]:
+                      - generic "Line number 1" [ref=e138]: "1"
+                      - generic [ref=e139]:
+                        - button "Collapse" [ref=e140] [cursor=pointer]: ▼
+                        - generic [ref=e141]: "{"
+                    - generic [ref=e142]:
+                      - generic [ref=e144]:
+                        - generic "Line number 2" [ref=e145]: "2"
+                        - generic [ref=e146]:
+                          - generic [ref=e147]: "\"repo_id\""
+                          - generic [ref=e148]: ":"
+                          - generic [ref=e149]: "\"stabilityai/stable-audio-3-small-sfx\""
+                          - generic [ref=e150]: ","
+                      - generic [ref=e152]:
+                        - generic "Line number 3" [ref=e153]: "3"
+                        - generic [ref=e154]:
+                          - generic [ref=e155]: "\"family\""
+                          - generic [ref=e156]: ":"
+                          - generic [ref=e157]: "\"post-trained\""
+                          - generic [ref=e158]: ","
+                      - generic [ref=e160]:
+                        - generic "Line number 4" [ref=e161]: "4"
+                        - generic [ref=e162]:
+                          - generic [ref=e163]: "\"note\""
+                          - generic [ref=e164]: ":"
+                          - generic [ref=e165]: "\"Lightweight sound-effects checkpoint.\""
+                          - generic [ref=e166]: ","
+                      - generic [ref=e168]:
+                        - generic "Line number 5" [ref=e169]: "5"
+                        - generic [ref=e170]:
+                          - generic [ref=e171]: "\"token_hint\""
+                          - generic [ref=e172]: ":"
+                          - generic [ref=e173]: "\"This is a gated Stability model. Accept the model terms on Hugging Face and add a read-only HF_TOKEN Space secret if download fails.\""
+                      - generic [ref=e174]:
+                        - generic "Line number 6" [ref=e175]: "6"
+                        - generic [ref=e177]: "}"
+                - generic [ref=e178]:
+                  - generic:
+                    - generic:
+                      - img
+                    - text: Output
+                  - generic "Empty value" [ref=e179]:
+                    - img [ref=e181]
+                - generic [ref=e185]:
+                  - generic [ref=e186]:
+                    - generic:
+                      - generic:
+                        - img
+                      - text: Run metadata
+                  - button "Copy" [ref=e215] [cursor=pointer]:
+                    - img [ref=e217]
+                  - generic [ref=e221]:
+                    - generic [ref=e222]:
+                      - generic "Line number 1" [ref=e223]: "1"
+                      - generic [ref=e224]:
+                        - button "Collapse" [ref=e225] [cursor=pointer]: ▼
+                        - generic [ref=e226]: "{"
+                    - generic [ref=e227]:
+                      - generic [ref=e229]:
+                        - generic "Line number 2" [ref=e230]: "2"
+                        - generic [ref=e231]:
+                          - generic [ref=e232]: "\"device\""
+                          - generic [ref=e233]: ":"
+                          - generic [ref=e234]: "\"cpu\""
+                          - generic [ref=e235]: ","
+                      - generic [ref=e237]:
+                        - generic "Line number 3" [ref=e238]: "3"
+                        - generic [ref=e239]:
+                          - generic [ref=e240]: "\"cuda_name\""
+                          - generic [ref=e241]: ":"
+                          - generic [ref=e242]: "null"
+                          - generic [ref=e243]: ","
+                      - generic [ref=e245]:
+                        - generic "Line number 4" [ref=e246]: "4"
+                        - generic [ref=e247]:
+                          - generic [ref=e248]: "\"flash_attn\""
+                          - generic [ref=e249]: ":"
+                          - generic [ref=e250]: "true"
+                          - generic [ref=e251]: ","
+                      - generic [ref=e253]:
+                        - generic "Line number 5" [ref=e254]: "5"
+                        - generic [ref=e255]:
+                          - generic [ref=e256]: "\"hf_token_present\""
+                          - generic [ref=e257]: ":"
+                          - generic [ref=e258]: "false"
+                          - generic [ref=e259]: ","
+                      - generic [ref=e261]:
+                        - generic "Line number 6" [ref=e262]: "6"
+                        - generic [ref=e263]:
+                          - generic [ref=e264]: "\"loaded_generation_model\""
+                          - generic [ref=e265]: ":"
+                          - generic [ref=e266]: "null"
+                          - generic [ref=e267]: ","
+                      - generic [ref=e269]:
+                        - generic "Line number 7" [ref=e270]: "7"
+                        - generic [ref=e271]:
+                          - generic [ref=e272]: "\"loaded_autoencoder\""
+                          - generic [ref=e273]: ":"
+                          - generic [ref=e274]: "null"
+                      - generic [ref=e275]:
+                        - generic "Line number 8" [ref=e276]: "8"
+                        - generic [ref=e278]: "}"
+      - generic [ref=e192]:
+        - button "Use via API logo" [ref=e193] [cursor=pointer]:
+          - text: Use via API
+          - img "logo" [ref=e194]
+        - generic [ref=e195]: ·
+        - link "Built with Gradio logo" [ref=e196] [cursor=pointer]:
+          - /url: https://gradio.app
+          - text: Built with Gradio
+          - img "logo" [ref=e197]
+        - generic [ref=e198]: ·
+        - button "Settings Settings" [ref=e199] [cursor=pointer]:
+          - text: Settings
+          - img "Settings" [ref=e200]
+  - generic [ref=e201]:
+    - generic [ref=e202]:
+      - img [ref=e203]
+      - link "owenisas" [ref=e204] [cursor=pointer]:
+        - /url: https://huggingface.co/owenisas
+      - generic [ref=e205]: /
+      - link "stable-audio-3-lab" [ref=e206] [cursor=pointer]:
+        - /url: https://huggingface.co/spaces/owenisas/stable-audio-3-lab
+      - link "0" [ref=e207] [cursor=pointer]:
+        - /url: https://huggingface.co/spaces/owenisas/stable-audio-3-lab
+        - img [ref=e208]
+        - paragraph [ref=e210]: "0"
+    - img [ref=e212] [cursor=pointer]

.playwright-cli/page-2026-05-20T20-24-02-635Z.yml

@@ -0,0 +1,224 @@
+- generic [ref=e1]:
+  - main [ref=e4]:
+    - generic [ref=e21]:
+      - generic [ref=e23]:
+        - heading "Stable Audio 3 Lab" [level=1] [ref=e28]
+        - generic [ref=e29]:
+          - generic [ref=e30]:
+            - generic [ref=e31]:
+              - button [ref=e32] [cursor=pointer]: Generate
+              - button [ref=e33] [cursor=pointer]: Autoencoder
+              - button [ref=e34] [cursor=pointer]: Coverage
+            - tablist [ref=e35]:
+              - tab "Generate" [selected] [ref=e36] [cursor=pointer]
+              - tab "Autoencoder" [ref=e37] [cursor=pointer]
+              - tab "Coverage" [ref=e38] [cursor=pointer]
+          - tabpanel [ref=e39]:
+            - generic [ref=e41]:
+              - generic [ref=e42]:
+                - generic [ref=e43]:
+                  - generic [ref=e45]:
+                    - generic [ref=e46]: Model
+                    - generic [ref=e49]:
+                      - listbox "Model" [ref=e50]: Stable Audio 3 Small SFX
+                      - generic:
+                        - img
+                  - generic [ref=e52]:
+                    - generic [ref=e53]: Prompt
+                    - textbox "Prompt" [ref=e55]:
+                      - /placeholder: ""
+                      - text: Close binaural rain on a window, soft cloth movement, detailed texture
+                  - generic [ref=e57]:
+                    - generic [ref=e58]: Negative prompt
+                    - textbox "Negative prompt" [ref=e60]:
+                      - /placeholder: ""
+                - generic [ref=e62]:
+                  - generic [ref=e64]:
+                    - generic [ref=e65]:
+                      - generic [ref=e67]: Duration
+                      - generic [ref=e68]:
+                        - spinbutton "number input for Duration" [ref=e69]: "8"
+                        - button "Reset to default value" [ref=e70] [cursor=pointer]: ↺
+                    - generic [ref=e71]:
+                      - generic [ref=e72]: "1"
+                      - slider "range slider for Duration" [ref=e73] [cursor=pointer]: "8"
+                      - generic [ref=e74]: "120"
+                  - generic [ref=e76]:
+                    - generic [ref=e77]:
+                      - generic [ref=e79]: Steps
+                      - generic [ref=e80]:
+                        - spinbutton "number input for Steps" [ref=e81]: "8"
+                        - button "Reset to default value" [ref=e82] [cursor=pointer]: ↺
+                    - generic [ref=e83]:
+                      - generic [ref=e84]: "1"
+                      - slider "range slider for Steps" [ref=e85] [cursor=pointer]: "8"
+                      - generic [ref=e86]: "100"
+                  - generic [ref=e88]:
+                    - generic [ref=e89]:
+                      - generic [ref=e91]: CFG
+                      - generic [ref=e92]:
+                        - spinbutton "number input for CFG" [ref=e93]: "1"
+                        - button "Reset to default value" [ref=e94] [cursor=pointer]: ↺
+                    - generic [ref=e95]:
+                      - generic [ref=e96]: "0"
+                      - slider "range slider for CFG" [ref=e97] [cursor=pointer]: "1"
+                      - generic [ref=e98]: "12"
+                - generic [ref=e100]:
+                  - generic [ref=e102]:
+                    - generic [ref=e103]: Sampler
+                    - generic [ref=e106]:
+                      - listbox "Sampler" [ref=e107]: pingpong
+                      - generic:
+                        - img
+                  - generic [ref=e109]:
+                    - generic [ref=e110]: Seed
+                    - spinbutton "Seed" [ref=e111]: "-1"
+                - generic [ref=e113]:
+                  - generic [ref=e115] [cursor=pointer]:
+                    - checkbox "Chunked decode" [checked] [ref=e116]
+                    - generic [ref=e117]: Chunked decode
+                  - generic [ref=e119] [cursor=pointer]:
+                    - checkbox "CPU override" [ref=e120]
+                    - generic [ref=e121]: CPU override
+                - generic [ref=e122]:
+                  - button "Generate" [active] [ref=e123] [cursor=pointer]
+                  - button "Unload" [ref=e124] [cursor=pointer]
+                  - button "Runtime" [ref=e125] [cursor=pointer]
+              - generic [ref=e126]:
+                - generic [ref=e127]:
+                  - generic [ref=e128]:
+                    - generic:
+                      - generic:
+                        - img
+                      - text: Model info
+                  - button "Copy" [ref=e130] [cursor=pointer]:
+                    - img [ref=e132]
+                  - generic [ref=e136]:
+                    - generic [ref=e137]:
+                      - generic "Line number 1" [ref=e138]: "1"
+                      - generic [ref=e139]:
+                        - button "Collapse" [ref=e140] [cursor=pointer]: ▼
+                        - generic [ref=e141]: "{"
+                    - generic [ref=e142]:
+                      - generic [ref=e144]:
+                        - generic "Line number 2" [ref=e145]: "2"
+                        - generic [ref=e146]:
+                          - generic [ref=e147]: "\"repo_id\""
+                          - generic [ref=e148]: ":"
+                          - generic [ref=e149]: "\"stabilityai/stable-audio-3-small-sfx\""
+                          - generic [ref=e150]: ","
+                      - generic [ref=e152]:
+                        - generic "Line number 3" [ref=e153]: "3"
+                        - generic [ref=e154]:
+                          - generic [ref=e155]: "\"family\""
+                          - generic [ref=e156]: ":"
+                          - generic [ref=e157]: "\"post-trained\""
+                          - generic [ref=e158]: ","
+                      - generic [ref=e160]:
+                        - generic "Line number 4" [ref=e161]: "4"
+                        - generic [ref=e162]:
+                          - generic [ref=e163]: "\"note\""
+                          - generic [ref=e164]: ":"
+                          - generic [ref=e165]: "\"Lightweight sound-effects checkpoint.\""
+                          - generic [ref=e166]: ","
+                      - generic [ref=e168]:
+                        - generic "Line number 5" [ref=e169]: "5"
+                        - generic [ref=e170]:
+                          - generic [ref=e171]: "\"token_hint\""
+                          - generic [ref=e172]: ":"
+                          - generic [ref=e173]: "\"This is a gated Stability model. Accept the model terms on Hugging Face and add a read-only HF_TOKEN Space secret if download fails.\""
+                      - generic [ref=e174]:
+                        - generic "Line number 6" [ref=e175]: "6"
+                        - generic [ref=e177]: "}"
+                - generic [ref=e178]:
+                  - generic:
+                    - generic:
+                      - img
+                    - text: Output
+                  - generic "Empty value" [ref=e179]:
+                    - img [ref=e181]
+                - generic [ref=e185]:
+                  - generic [ref=e186]:
+                    - generic:
+                      - generic:
+                        - img
+                      - text: Run metadata
+                  - button "Copy" [ref=e215] [cursor=pointer]:
+                    - img [ref=e217]
+                  - generic [ref=e221]:
+                    - generic [ref=e222]:
+                      - generic "Line number 1" [ref=e223]: "1"
+                      - generic [ref=e224]:
+                        - button "Collapse" [ref=e225] [cursor=pointer]: ▼
+                        - generic [ref=e226]: "{"
+                    - generic [ref=e227]:
+                      - generic [ref=e229]:
+                        - generic "Line number 2" [ref=e230]: "2"
+                        - generic [ref=e231]:
+                          - generic [ref=e232]: "\"status\""
+                          - generic [ref=e233]: ":"
+                          - generic [ref=e234]: "\"blocked\""
+                          - generic [ref=e235]: ","
+                      - generic [ref=e237]:
+                        - generic "Line number 3" [ref=e238]: "3"
+                        - generic [ref=e239]:
+                          - generic [ref=e240]: "\"error\""
+                          - generic [ref=e241]: ":"
+                          - generic [ref=e279]: "\"Stable Audio 3 Small SFX is gated. Accept the model terms on Hugging Face, then add a read-only HF_TOKEN secret to this Space before running it.\""
+                          - generic [ref=e243]: ","
+                      - generic [ref=e245]:
+                        - generic "Line number 4" [ref=e246]: "4"
+                        - generic [ref=e247]:
+                          - generic [ref=e248]: "\"model\""
+                          - generic [ref=e249]: ":"
+                          - generic [ref=e280]: "\"small-sfx\""
+                          - generic [ref=e251]: ","
+                      - generic [ref=e253]:
+                        - generic "Line number 5" [ref=e254]: "5"
+                        - generic [ref=e255]:
+                          - generic [ref=e256]: "\"repo_id\""
+                          - generic [ref=e257]: ":"
+                          - generic [ref=e281]: "\"stabilityai/stable-audio-3-small-sfx\""
+                          - generic [ref=e259]: ","
+                      - generic [ref=e261]:
+                        - generic "Line number 6" [ref=e262]: "6"
+                        - generic [ref=e263]:
+                          - generic [ref=e264]: "\"device\""
+                          - generic [ref=e265]: ":"
+                          - generic [ref=e282]: "\"unknown\""
+                          - generic [ref=e267]: ","
+                      - generic [ref=e269]:
+                        - generic "Line number 7" [ref=e270]: "7"
+                        - generic [ref=e271]:
+                          - generic [ref=e272]: "\"hf_token_present\""
+                          - generic [ref=e273]: ":"
+                          - generic [ref=e283]: "false"
+                      - generic [ref=e275]:
+                        - generic "Line number 8" [ref=e276]: "8"
+                        - generic [ref=e278]: "}"
+      - generic [ref=e192]:
+        - button "Use via API logo" [ref=e193] [cursor=pointer]:
+          - text: Use via API
+          - img "logo" [ref=e194]
+        - generic [ref=e195]: ·
+        - link "Built with Gradio logo" [ref=e196] [cursor=pointer]:
+          - /url: https://gradio.app
+          - text: Built with Gradio
+          - img "logo" [ref=e197]
+        - generic [ref=e198]: ·
+        - button "Settings Settings" [ref=e199] [cursor=pointer]:
+          - text: Settings
+          - img "Settings" [ref=e200]
+  - generic [ref=e201]:
+    - generic [ref=e202]:
+      - img [ref=e203]
+      - link "owenisas" [ref=e204] [cursor=pointer]:
+        - /url: https://huggingface.co/owenisas
+      - generic [ref=e205]: /
+      - link "stable-audio-3-lab" [ref=e206] [cursor=pointer]:
+        - /url: https://huggingface.co/spaces/owenisas/stable-audio-3-lab
+      - link "0" [ref=e207] [cursor=pointer]:
+        - /url: https://huggingface.co/spaces/owenisas/stable-audio-3-lab
+        - img [ref=e208]
+        - paragraph [ref=e210]: "0"
+    - img [ref=e212] [cursor=pointer]

README.md

@@ -9,6 +9,9 @@ python_version: "3.10"
 suggested_hardware: a10g-small
 pinned: false
 license: mit
+hf_oauth: true
+hf_oauth_scopes:
+  - gated-repos
 ---
 
 # Stable Audio 3 Lab
@@ -23,10 +26,11 @@ The optimized repo (`stabilityai/stable-audio-3-optimized`) currently ships MLX
 
 ## Access
 
-The post-trained Stable Audio 3 checkpoints are gated on Hugging Face. Before using them here:
+This Space requires Hugging Face login. The post-trained Stable Audio 3
+checkpoints are gated on Hugging Face, so each user must:
 
-1. Accept the terms on each gated model page while logged in.
-2. Add a read-only `HF_TOKEN` secret to this Space.
+1. Sign in with Hugging Face.
+2. Accept the terms on each gated model page from their own account.
 
 Base checkpoints are not gated, but they are intended mainly for fine-tuning and may not sound as polished.
 

app.py

@@ -7,7 +7,11 @@ import json
 import os
 import sys
 import tempfile
+import threading
 import time
+import urllib.error
+import urllib.request
+from contextlib import contextmanager
 from dataclasses import dataclass
 from typing import Any
 
@@ -167,6 +171,7 @@ COLLECTION_ROWS = [
 
 MODEL_CACHE: dict[str, Any] = {"key": None, "model": None}
 AE_CACHE: dict[str, Any] = {"key": None, "model": None}
+MODEL_LOAD_LOCK = threading.RLock()
 
 
 def gpu_task(duration: int):
@@ -196,27 +201,78 @@ def flash_attn_available() -> bool:
     return importlib.util.find_spec("flash_attn") is not None
 
 
-def hf_token_present() -> bool:
-    return bool(os.getenv("HF_TOKEN") or os.getenv("HUGGING_FACE_HUB_TOKEN"))
+def oauth_token_value(oauth_token: gr.OAuthToken | None) -> str | None:
+    token = getattr(oauth_token, "token", None)
+    return token if isinstance(token, str) and token else None
+
+
+def oauth_username(oauth_profile: gr.OAuthProfile | None) -> str | None:
+    username = getattr(oauth_profile, "username", None)
+    return username if isinstance(username, str) and username else None
 
 
 def stable_audio_token_hint(model: GenerationModel) -> str:
     if not model.gated:
-        return ""
-    if hf_token_present():
-        return ""
+        return "Sign in with Hugging Face before running this Space."
     return (
-        "This is a gated Stability model. Accept the model terms on Hugging Face "
-        "and add a read-only HF_TOKEN Space secret if download fails."
+        "Sign in with Hugging Face and accept this gated model's terms from your "
+        "own account before running it."
+    )
+
+
+def user_can_download_gated_model(repo_id: str, token: str) -> tuple[bool, str | None]:
+    request = urllib.request.Request(
+        f"https://huggingface.co/{repo_id}/resolve/main/model_config.json",
+        method="HEAD",
+        headers={"Authorization": f"Bearer {token}"},
     )
+    try:
+        with urllib.request.urlopen(request, timeout=20) as response:
+            return response.status < 400, None
+    except urllib.error.HTTPError as exc:
+        if exc.code in {401, 403}:
+            return (
+                False,
+                "Your Hugging Face account does not have access to this gated model yet. "
+                "Open the model page while logged in, accept Stability's terms, then retry.",
+            )
+        return False, f"Hugging Face access check failed with HTTP {exc.code}."
+    except Exception as exc:
+        return False, f"Hugging Face access check failed: {exc!r}"
+
+
+@contextmanager
+def hub_download_token(token: str | None):
+    if not token:
+        yield
+        return
 
+    import stable_audio_3.model_configs as model_configs
 
-def generation_preflight_error(model: GenerationModel, allow_cpu_medium: bool) -> tuple[str | None, str]:
+    original_download = model_configs.hf_hub_download
+
+    def download_with_user_token(*args, **kwargs):
+        kwargs.setdefault("token", token)
+        return original_download(*args, **kwargs)
+
+    model_configs.hf_hub_download = download_with_user_token
+    try:
+        yield
+    finally:
+        model_configs.hf_hub_download = original_download
+
+
+def generation_preflight_error(
+    model: GenerationModel,
+    allow_cpu_medium: bool,
+    oauth_profile: gr.OAuthProfile | None,
+    oauth_token: gr.OAuthToken | None,
+) -> tuple[str | None, str]:
     device = "unknown"
-    if model.gated and not hf_token_present():
+    token = oauth_token_value(oauth_token)
+    if oauth_profile is None or not token:
         return (
-            f"{model.label} is gated. Accept the model terms on Hugging Face, "
-            "then add a read-only HF_TOKEN secret to this Space before running it.",
+            "Sign in with Hugging Face before running this Space.",
             device,
         )
 
@@ -228,6 +284,10 @@ def generation_preflight_error(model: GenerationModel, allow_cpu_medium: bool) -
             "Use a GPU Space or enable the CPU override for a slow/debug-only attempt.",
             device,
         )
+    if model.gated:
+        has_access, error = user_can_download_gated_model(model.repo_id, token)
+        if not has_access:
+            return error or "Your Hugging Face account cannot access this gated model.", device
     if model.requires_cuda and device == "cuda" and not flash_attn_available():
         return (
             f"{model.label} expects flash-attn on CUDA. Rebuild the Space with the "
@@ -237,8 +297,13 @@ def generation_preflight_error(model: GenerationModel, allow_cpu_medium: bool) -
     return None, device
 
 
-def assert_generation_runtime(model: GenerationModel, allow_cpu_medium: bool) -> str:
-    error, device = generation_preflight_error(model, allow_cpu_medium)
+def assert_generation_runtime(
+    model: GenerationModel,
+    allow_cpu_medium: bool,
+    oauth_profile: gr.OAuthProfile | None,
+    oauth_token: gr.OAuthToken | None,
+) -> str:
+    error, device = generation_preflight_error(model, allow_cpu_medium, oauth_profile, oauth_token)
     if error:
         raise gr.Error(error)
     return device
@@ -270,27 +335,45 @@ def clear_torch_memory() -> None:
     gc.collect()
 
 
-def load_generation_model(model_key: str, allow_cpu_medium: bool):
+def load_generation_model(
+    model_key: str,
+    allow_cpu_medium: bool,
+    oauth_profile: gr.OAuthProfile | None,
+    oauth_token: gr.OAuthToken | None,
+):
     model_def = GENERATION_MODELS[model_key]
-    device = assert_generation_runtime(model_def, allow_cpu_medium)
+    device = assert_generation_runtime(model_def, allow_cpu_medium, oauth_profile, oauth_token)
 
     if MODEL_CACHE["key"] == model_key and MODEL_CACHE["model"] is not None:
         return MODEL_CACHE["model"], device
 
-    MODEL_CACHE["model"] = None
-    MODEL_CACHE["key"] = None
-    clear_torch_memory()
+    with MODEL_LOAD_LOCK:
+        if MODEL_CACHE["key"] == model_key and MODEL_CACHE["model"] is not None:
+            return MODEL_CACHE["model"], device
+
+        MODEL_CACHE["model"] = None
+        MODEL_CACHE["key"] = None
+        clear_torch_memory()
+
+        from stable_audio_3 import StableAudioModel
 
-    from stable_audio_3 import StableAudioModel
+        model_half = device == "cuda"
+        with hub_download_token(oauth_token_value(oauth_token)):
+            model = StableAudioModel.from_pretrained(model_key, model_half=model_half)
+        MODEL_CACHE["key"] = model_key
+        MODEL_CACHE["model"] = model
+        return model, device
 
-    model_half = device == "cuda"
-    model = StableAudioModel.from_pretrained(model_key, model_half=model_half)
-    MODEL_CACHE["key"] = model_key
-    MODEL_CACHE["model"] = model
-    return model, device
 
+def load_autoencoder(
+    model_key: str,
+    allow_cpu_same_l: bool,
+    oauth_profile: gr.OAuthProfile | None,
+    oauth_token: gr.OAuthToken | None,
+):
+    if oauth_profile is None or not oauth_token_value(oauth_token):
+        raise gr.Error("Sign in with Hugging Face before running this Space.")
 
-def load_autoencoder(model_key: str, allow_cpu_same_l: bool):
     model_def = AUTOENCODER_MODELS[model_key]
     torch = import_torch()
     device = current_device(torch)
@@ -303,16 +386,21 @@ def load_autoencoder(model_key: str, allow_cpu_same_l: bool):
     if AE_CACHE["key"] == model_key and AE_CACHE["model"] is not None:
         return AE_CACHE["model"], device
 
-    AE_CACHE["model"] = None
-    AE_CACHE["key"] = None
-    clear_torch_memory()
+    with MODEL_LOAD_LOCK:
+        if AE_CACHE["key"] == model_key and AE_CACHE["model"] is not None:
+            return AE_CACHE["model"], device
+
+        AE_CACHE["model"] = None
+        AE_CACHE["key"] = None
+        clear_torch_memory()
 
-    from stable_audio_3 import AutoencoderModel
+        from stable_audio_3 import AutoencoderModel
 
-    model = AutoencoderModel.from_pretrained(model_key)
-    AE_CACHE["key"] = model_key
-    AE_CACHE["model"] = model
-    return model, device
+        with hub_download_token(oauth_token_value(oauth_token)):
+            model = AutoencoderModel.from_pretrained(model_key)
+        AE_CACHE["key"] = model_key
+        AE_CACHE["model"] = model
+        return model, device
 
 
 def model_changed(model_key: str):
@@ -346,6 +434,8 @@ def generate_audio(
     seed: int,
     chunked_decode: bool,
     allow_cpu_medium: bool,
+    oauth_profile: gr.OAuthProfile | None = None,
+    oauth_token: gr.OAuthToken | None = None,
     progress=gr.Progress(track_tqdm=True),
 ):
     model_def = GENERATION_MODELS[model_key]
@@ -357,7 +447,12 @@ def generate_audio(
             "repo_id": model_def.repo_id,
         }
 
-    preflight_error, preflight_device = generation_preflight_error(model_def, allow_cpu_medium)
+    preflight_error, preflight_device = generation_preflight_error(
+        model_def,
+        allow_cpu_medium,
+        oauth_profile,
+        oauth_token,
+    )
     if preflight_error:
         return None, {
             "status": "blocked",
@@ -365,7 +460,9 @@ def generate_audio(
             "model": model_def.key,
             "repo_id": model_def.repo_id,
             "device": preflight_device,
-            "hf_token_present": hf_token_present(),
+            "signed_in": oauth_profile is not None,
+            "username": oauth_username(oauth_profile),
+            "oauth_token_present": bool(oauth_token_value(oauth_token)),
         }
 
     progress(0.05, desc="Loading model")
@@ -374,7 +471,7 @@ def generate_audio(
     if seed < 0:
         seed = int.from_bytes(os.urandom(4), "little") % 100000
 
-    model, device = load_generation_model(model_key, allow_cpu_medium)
+    model, device = load_generation_model(model_key, allow_cpu_medium, oauth_profile, oauth_token)
     progress(0.25, desc="Generating")
     audio = model.generate(
         prompt=prompt.strip(),
@@ -412,6 +509,7 @@ def generate_audio(
         "elapsed_s": elapsed,
         "output_file": out_file.name,
         "note": model_def.note,
+        "username": oauth_username(oauth_profile),
     }
     return out_file.name, metadata
 
@@ -422,8 +520,19 @@ def roundtrip_autoencoder(
     audio_input: tuple[int, np.ndarray] | None,
     chunked: bool,
     allow_cpu_same_l: bool,
+    oauth_profile: gr.OAuthProfile | None = None,
+    oauth_token: gr.OAuthToken | None = None,
     progress=gr.Progress(track_tqdm=True),
 ):
+    if oauth_profile is None or not oauth_token_value(oauth_token):
+        return None, {
+            "status": "blocked",
+            "error": "Sign in with Hugging Face before running this Space.",
+            "autoencoder": model_key,
+            "repo_id": AUTOENCODER_MODELS[model_key]["repo_id"],
+            "signed_in": oauth_profile is not None,
+        }
+
     if audio_input is None:
         return None, {
             "status": "blocked",
@@ -449,7 +558,7 @@ def roundtrip_autoencoder(
 
     progress(0.05, desc="Loading autoencoder")
     started = time.time()
-    model, device = load_autoencoder(model_key, allow_cpu_same_l)
+    model, device = load_autoencoder(model_key, allow_cpu_same_l, oauth_profile, oauth_token)
 
     progress(0.25, desc="Encoding")
     sr, data = audio_input
@@ -480,20 +589,26 @@ def roundtrip_autoencoder(
         "latent_shape": list(latents.shape),
         "elapsed_s": round(time.time() - started, 3),
         "output_file": out_file.name,
+        "username": oauth_username(oauth_profile),
     }
     return out_file.name, metadata
 
 
-def unload_models():
+def unload_models(oauth_profile: gr.OAuthProfile | None = None):
+    if oauth_profile is None:
+        return {"status": "blocked", "error": "Sign in with Hugging Face before running this Space."}
     MODEL_CACHE["key"] = None
     MODEL_CACHE["model"] = None
     AE_CACHE["key"] = None
     AE_CACHE["model"] = None
     clear_torch_memory()
-    return {"status": "unloaded"}
+    return {"status": "unloaded", "username": oauth_username(oauth_profile)}
 
 
-def runtime_status():
+def runtime_status(
+    oauth_profile: gr.OAuthProfile | None = None,
+    oauth_token: gr.OAuthToken | None = None,
+):
     try:
         torch = import_torch()
         device = current_device(torch)
@@ -507,7 +622,9 @@ def runtime_status():
         "device": device,
         "cuda_name": cuda_name,
         "flash_attn": flash_attn_available(),
-        "hf_token_present": hf_token_present(),
+        "signed_in": oauth_profile is not None,
+        "username": oauth_username(oauth_profile),
+        "oauth_token_present": bool(oauth_token_value(oauth_token)),
         "loaded_generation_model": MODEL_CACHE["key"],
         "loaded_autoencoder": AE_CACHE["key"],
     }
@@ -524,6 +641,7 @@ css = """
 
 with gr.Blocks(title="Stable Audio 3 Lab") as demo:
     gr.Markdown("# Stable Audio 3 Lab")
+    gr.LoginButton(value="Sign in with Hugging Face", logout_value="Logout ({})")
 
     with gr.Tab("Generate"):
         with gr.Row(equal_height=False):