Load the AoTI packages from the public multimodalart/minimax-h3-aoti, move the upsample toggle under the prompt

README.md

@@ -107,7 +107,7 @@ canvas and frame count rather than re-deriving them.
 ## AoTI-compiled blocks
 
 With `H3_AOTI=1` the 50 repeated transformer blocks run from a compiled package,
-`diffusers-internal-dev/minimax-h3-aoti:bf16/torch2.11/sm120/dynamic` — a single dynamic-sequence artifact that serves
+[`multimodalart/minimax-h3-aoti`](https://huggingface.co/multimodalart/minimax-h3-aoti)`:bf16/torch2.11/sm120/dynamic` — a single dynamic-sequence artifact that serves
 every canvas, duration, reference set and prompt length.
 
 It is the **same package the `transformer/` partition runs**, and nothing about it is partition-specific. The two
@@ -183,10 +183,11 @@ on is cold and a cold one pays the lazy 72.16 GiB `PIPE.to("cuda")` inside its f
 
 ## Secrets
 
-The weights are the public [`MiniMaxAI/MiniMax-H3`](https://huggingface.co/MiniMaxAI/MiniMax-H3) diffusers
-checkpoint and the conditioner is a public Space called without a token, so that round trip runs on the caller's own
-quota rather than this org's. `HF_TOKEN` is still read by `h3_aoti`, whose compiled-package repository is private —
-without it, set `H3_AOTI=0`.
+Nothing this Space loads is private any more: the weights are the public
+[`MiniMaxAI/MiniMax-H3`](https://huggingface.co/MiniMaxAI/MiniMax-H3) checkpoint, the compiled AoTI packages are the
+public [`multimodalart/minimax-h3-aoti`](https://huggingface.co/multimodalart/minimax-h3-aoti) model repo, and the
+conditioner is a public Space called without a token — so that round trip runs on the caller's own quota rather than
+this org's. No `HF_TOKEN` is required.
 
 ## Where diffusers comes from
 

app.py

@@ -237,8 +237,8 @@ def load_models() -> str | None:
         blocks = MiniMaxH3Ref2VAGeneratorBlocks()
         print(f"[ref2va] loading {[c.name for c in blocks.expected_components]} from {MODEL_REPO} ...", flush=True)
         pipe = blocks.init_pipeline(MODEL_REPO, components_manager=manager, collection="h3")
-        # `MiniMaxAI/MiniMax-H3` is public, so the weights are fetched without a token. `HF_TOKEN` is still read by
-        # `h3_aoti`, whose compiled-package repository is not.
+        # Every repository this Space reads is public — the checkpoint, the AoTI packages and the conditioner
+        # Space — so no token is passed anywhere.
         pipe.load_components(dtype=torch.bfloat16)
 
         # Pin the two autoencoders to torch SDPA *before* the transformer takes cuDNN, and in that order.
@@ -552,7 +552,7 @@ def generate(
         f"({generate_seconds / int(steps):.1f} s/step) · seed {int(seed)}",
         flush=True,
     )
-    return path, refined
+    return path, refined, gr.update(visible=bool(refined))
 
 
 load_models()
@@ -585,6 +585,7 @@ with gr.Blocks(title="MiniMax-H3 Reference") as demo:
                 lines=3,
                 value="The character walks through a neon-lit street in the rain, humming to themselves",
             )
+            upsample = gr.Checkbox(label="Upsample prompt", value=False)
             # One tab per modality, in the order the model reads them. A reference left in a tab that is not the
             # open one is still part of the request — the tabs lay the slots out, they do not choose between them.
             with gr.Tabs():
@@ -618,21 +619,13 @@ with gr.Blocks(title="MiniMax-H3 Reference") as demo:
                 )
                 steps = gr.Slider(label="Steps", minimum=10, maximum=40, step=1, value=28)
                 seed = gr.Number(label="Seed", value=42, precision=0)
-                upsample = gr.Checkbox(
-                    label="Upsample prompt",
-                    value=False,
-                    info="Rewrites the prompt into the model's trained format with the conditioner's Qwen3-VL before encoding.",
-                )
 
         with gr.Column():
             result = gr.Video(label="Video + soundtrack")
-            with gr.Accordion("Upsampled prompt", open=False):
-                upsampled = gr.Textbox(
-                    show_label=False,
-                    lines=8,
-                    interactive=False,
-                    placeholder="Turn on “Upsample prompt” to see the rewrite that was encoded.",
-                )
+            # Only shown for a request that actually asked for a rewrite, so a plain request is not left with an
+            # empty panel. The accordion is an output for that reason: its visibility is part of the answer.
+            with gr.Accordion("Upsampled prompt", open=False, visible=False) as upsampled_panel:
+                upsampled = gr.Textbox(show_label=False, lines=8, interactive=False)
 
     open_slots = gr.State(OPEN_IMAGE_SLOTS)
 
@@ -681,14 +674,14 @@ with gr.Blocks(title="MiniMax-H3 Reference") as demo:
             ],
         ],
         inputs=[prompt, images[0], audio, video, canvas],
-        outputs=[result, upsampled],
+        outputs=[result, upsampled, upsampled_panel],
         fn=generate,
         cache_examples=True,
         cache_mode="lazy",
     )
 
     # The video stays the first output and the upsampled prompt is appended last, so existing consumers are untouched.
-    run.click(generate, request, [result, upsampled], api_name="generate")
+    run.click(generate, request, [result, upsampled, upsampled_panel], api_name="generate")
 
 
 if __name__ == "__main__":

h3_aoti.py

@@ -2,7 +2,7 @@
 
 Shared byte-identically by every MiniMax-H3 Space. A Space only ever calls `maybe_load()`; the compile path runs from
 the debug Space's "Compile (AoTI)" tab, or off-Space from `job_bf16_aoti.py` on an `rtx-pro-6000` Job, and pushes its
-artifacts to `diffusers-internal-dev/minimax-h3-aoti` under `<width>/torch<X.Y>/sm<cc>/<shape>`.
+artifacts to `multimodalart/minimax-h3-aoti` under `<width>/torch<X.Y>/sm<cc>/<shape>`.
 
 What is measured, so nobody has to guess whether this is worth turning on. Unquantized bfloat16, 124 frames,
 everything resident, one dynamic-sequence package serving every row — on an RTX PRO 6000 Blackwell, torch 2.11,
@@ -62,8 +62,11 @@ import os
 from pathlib import Path
 
 AOTI = os.environ.get("H3_AOTI", "0") == "1"
-AOTI_REPO = os.environ.get("H3_AOTI_REPO", "diffusers-internal-dev/minimax-h3-aoti")
-AOTI_REPO_TYPE = os.environ.get("H3_AOTI_REPO_TYPE", "dataset")
+# A public **model** repo. It used to be a private dataset, which is why the repo type is still a variable: the
+# artifacts are keyed by quant/torch/arch under `<width>/torch<X.Y>/sm<cc>/<shape>` rather than laid out the way
+# `spaces.aoti_load` expects, so the download is done by hand either way (see `maybe_load`).
+AOTI_REPO = os.environ.get("H3_AOTI_REPO", "multimodalart/minimax-h3-aoti")
+AOTI_REPO_TYPE = os.environ.get("H3_AOTI_REPO_TYPE", "model")
 # `dynamic` is the one package that serves every canvas, duration *and prompt*, and for bfloat16 it is what gets built:
 # a dynamic sequence dimension exports and compiles cleanly (measured on an rtx-pro-6000 Job, torch 2.11). It has to be
 # dynamic to be useful at all — `build_packed_sequence` pads nothing, so
@@ -211,8 +214,8 @@ def maybe_load(transformer) -> None:
     """Patch the block stack with its compiled package. Once, and safe to call at **startup**.
 
     Nothing here touches a GPU: the download is CPU work and the `.pt2` archive is not opened until the first forward,
-    which happens inside the `@spaces.GPU` call. Proven on the pool — `diffusers-internal-dev/minimax-h3-generator-aoti`
-    loads `bf16/torch2.11/sm120/dynamic` at startup, patches all 50 blocks, and generates.
+    which happens inside the `@spaces.GPU` call. Proven on the pool: `bf16/torch2.11/sm120/dynamic` loads at startup,
+    patches all 50 blocks, and generates. The repo is public, so no token is passed for it.
     """
     if not AOTI or id(transformer) in _LOADED:
         return
@@ -226,7 +229,6 @@ def maybe_load(transformer) -> None:
         repo_id=AOTI_REPO,
         repo_type=AOTI_REPO_TYPE,
         allow_patterns=f"{key}/package/*",
-        token=os.environ.get("HF_TOKEN"),
     )
     package_dir = Path(local) / key / "package"
     if not package_dir.is_dir():
@@ -420,7 +422,7 @@ def upload(package_dir: str | os.PathLike[str], key: str) -> str:
     if not token:
         raise RuntimeError("`HF_TOKEN` is needed to push the AoTI package.")
     api = HfApi(token=token)
-    api.create_repo(repo_id=AOTI_REPO, repo_type=AOTI_REPO_TYPE, private=True, exist_ok=True)
+    api.create_repo(repo_id=AOTI_REPO, repo_type=AOTI_REPO_TYPE, private=False, exist_ok=True)
     api.upload_folder(
         folder_path=str(package_dir),
         path_in_repo=f"{key}/package",