Fix conditioner call in _generate; drop unsupported label on Markdown output

app.py

@@ -3,7 +3,7 @@
 Ported from [@mrfakename/minimax-h3-ultra-fast](https://huggingface.co/spaces/mrfakename/minimax-h3-ultra-fast),
 trimmed to the text-to-video path: the React studio, keyframe/FL2VA inputs, Ref2VA references, storyboard stitching
 and custom-LoRA URL downloads are removed, and the UI is an ordinary Gradio Blocks app. The engine is unchanged —
-a pruned NVFP4 transformer plus a local truncated NVFP4 conditioner, with the full-precision VAEs kept canonical.
+a pruned NVFP4 transformer plus a local truncated NVFP4 conditioner, with the full-precision VAEs kept.
 
 Env overrides (`H3_*`) behave exactly as in the source Space: `H3_ENGINE=nvfp4|bf16`, `H3_CONDITIONER_MODE=local|remote`,
 `H3_PLACEMENT=lazy|pack|offload`, `H3_ATTENTION`, `H3_GPU_SIZE`, `H3_MAX_GPU_DURATION`.
@@ -16,7 +16,6 @@ import os
 import re
 import tempfile
 import time
-import traceback
 from collections import OrderedDict
 from functools import cache
 
@@ -354,8 +353,6 @@ def _download_egrid() -> str:
         if _sha256(path) == EGRID_SHA256:
             return path
     temporary = path + ".download"
-    import requests
-
     with requests.get(EGRID_URL, stream=True, timeout=60) as response:
         response.raise_for_status()
         with open(temporary, "wb") as output:
@@ -389,7 +386,7 @@ def resolve_lora(preset: str) -> tuple[str | None, str | None, str, float]:
         raise ValueError("LoRA file is missing a declared size or exceeds the 2 GiB safety limit.")
     path = hf_hub_download(repo_id=repo_id, filename=filename, token=False)
     # Only the older pruned-base adapter contains AdaLN targets and needs its external timestep lookup grid.
-    egrid_path = _download_egrid() if needs_egrid else None
+    egrid_path = None if not needs_egrid else _download_egrid()
     return path, egrid_path, f"{repo_id}/{filename}", adapter_scale
 
 
@@ -454,7 +451,7 @@ def _generate(
             print(f"[cond] reused {num_text_tokens}-token embedding", flush=True)
         else:
             conditioned = time.time()
-            condition_state = active_conditioner(
+            condition_state = COND_PIPE(
                 prompt=prompt,
                 image=None,
                 last_image=None,
@@ -741,7 +738,7 @@ with demo:
             generate_btn = gr.Button("Generate", variant="primary")
         with gr.Column(scale=2):
             video_out = gr.Video(label="Result (video + audio)")
-            report = gr.Markdown(label="Run report")
+            report = gr.Markdown()
 
     generation_preset.change(
         lambda preset: gr.Accordion(visible=preset == CUSTOM_PRESET),