Fix: don't use rank-768 LoRA alongside full distil model; capture stderr

app.py

@@ -25,8 +25,7 @@ FINETUNES_DIR = WAN2GP_ROOT / "finetunes"
 GENERATE_PY  = Path(__file__).parent / "generate.py"
 
 SULPHUR_ASSETS = [
-    ("SulphurAI/Sulphur-2-base", "sulphur_distil_bf16.safetensors",  CKPTS_DIR),
-    ("SulphurAI/Sulphur-2-base", "sulphur_lora_rank_768.safetensors", LORAS_DIR),
+    ("SulphurAI/Sulphur-2-base", "sulphur_distil_bf16.safetensors", CKPTS_DIR),
 ]
 LTX_ASSETS = [
     ("DeepBeepMeep/LTX-2", "ltx-2.3-22b-distilled-lora-384-1.1.safetensors", LORAS_DIR),
@@ -42,8 +41,9 @@ SULPHUR_FINETUNE = {
         "architecture": "ltx2_22B",
         "parent_model_type": "ltx2_22B",
         "description": "LTX-2.3 fine-tuned i2v. Distilled checkpoint.",
+        # Full distilled model — do NOT also preload the rank-768 LoRA (README: use one or the other)
         "URLs": [str(CKPTS_DIR / "sulphur_distil_bf16.safetensors")],
-        "preload_URLs": [str(LORAS_DIR / "sulphur_lora_rank_768.safetensors") + "|%lora_dir"],
+        "preload_URLs": [],
     },
     "num_inference_steps": 8,
     "video_length": 81,
@@ -125,12 +125,17 @@ def _generate(image: str, prompt: str, model: str, steps: int,
     ]
 
     log_lines = []
-    proc = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT,
+    proc = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
                             text=True, bufsize=1, env=env)
     for line in proc.stdout:
         log_lines.append(line.rstrip())
         print(line, end="")
+    stderr_out = proc.stderr.read()
     proc.wait()
+    if stderr_out:
+        log_lines.append("--- stderr ---")
+        log_lines.extend(stderr_out.splitlines())
+        print(stderr_out)
 
     log = "\n".join(log_lines)
     if proc.returncode != 0 or not out_file.exists():