[pull] master from Comfy-Org:master

comfy/cli_args.py

@@ -172,6 +172,7 @@ def from_string(cls, value: str):
 
 parser.add_argument("--reserve-vram", type=float, default=None, help="Set the amount of vram in GB you want to reserve for use by your OS/other software. By default some amount is reserved depending on your OS.")
 parser.add_argument("--vram-headroom", type=float, default=0, help="Set the amount of vram in GB for DynamicVRAM to maintain as extra headroom above default. ComfyUI will try and keep this much VRAM completely free and unused, even counting VRAM from other apps.")
+parser.add_argument("--disable-nvml-pressure", action="store_true", help="Use CUDA instead of NVML for DynamicVRAM memory pressure.")
 
 parser.add_argument("--async-offload", nargs='?', const=2, type=int, default=None, metavar="NUM_STREAMS", help="Use async weight offloading. An optional argument controls the amount of offload streams. Default is 2. Enabled by default on Nvidia.")
 parser.add_argument("--disable-async-offload", action="store_true", help="Disable async weight offloading.")

comfy/model_management.py

@@ -671,6 +671,19 @@ def pin_eviction_tiers(loaded, evict_active):
             tiers.append((PIN_SUBSETS, True, True))
     return tiers
 
+def registration_eviction_tiers(evict_active):
+    subsets = PIN_SUBSETS + LOADED_PIN_SUBSETS
+    tiers = [
+        (subsets, False, False),
+        (subsets, True, False),
+    ]
+    if evict_active:
+        tiers.extend([
+            (subsets, False, True),
+            (subsets, True, True),
+        ])
+    return tiers
+
 def free_pins(size, evict_active=False, loaded=False):
     freed = 0
     for subsets, current_prompt, active in pin_eviction_tiers(loaded, evict_active):
@@ -703,19 +716,19 @@ def ensure_pin_budget(size, evict_active=False, loaded=False):
     to_free = shortfall + PIN_PRESSURE_HYSTERESIS
     return free_pins(to_free, evict_active=evict_active, loaded=loaded) >= shortfall
 
-def free_registrations(shortfall, evict_active=True, loaded=False):
+def free_registrations(shortfall, evict_active=True):
     if MAX_PINNED_MEMORY <= 0:
         return False
     if shortfall <= 0:
         return True
 
     shortfall += REGISTERABLE_PIN_HYSTERESIS
-    for subsets, current_prompt, active in pin_eviction_tiers(loaded, evict_active):
+    for subsets, current_prompt, active in registration_eviction_tiers(evict_active):
         shortfall -= free_model_pins(shortfall, subsets, current_prompt, active, registrations=True)
     return shortfall <= REGISTERABLE_PIN_HYSTERESIS
 
-def ensure_pin_registerable(size, evict_active=True, loaded=False):
-    return free_registrations(TOTAL_PINNED_MEMORY + size - MAX_PINNED_MEMORY, evict_active=evict_active, loaded=loaded)
+def ensure_pin_registerable(size, evict_active=True):
+    return free_registrations(TOTAL_PINNED_MEMORY + size - MAX_PINNED_MEMORY, evict_active=evict_active)
 
 class LoadedModel:
     def __init__(self, model: ModelPatcher):

comfy/pinned_memory.py

@@ -56,7 +56,7 @@ def get_pin(module, subset="weights"):
 
     _, _, stack_split, pinned_size, *_ = module._pin_state[subset]
     size = pin.nbytes
-    comfy.model_management.ensure_pin_registerable(size, loaded=subset.endswith("-loaded"))
+    comfy.model_management.ensure_pin_registerable(size)
 
     if torch.cuda.cudart().cudaHostRegister(pin.data_ptr(), size, 1) != 0:
         comfy.model_management.discard_cuda_async_error()
@@ -93,7 +93,7 @@ def pin_memory(module, subset="weights", size=None):
 
     comfy.memory_management.extra_ram_release(comfy.memory_management.RAM_CACHE_HEADROOM)
     if (not comfy.model_management.ensure_pin_budget(size, loaded=loaded) or
-        not comfy.model_management.ensure_pin_registerable(registerable_size, loaded=loaded)):
+        not comfy.model_management.ensure_pin_registerable(registerable_size)):
         return _steal_pin(module, stack, buckets, size, priority, subset)
 
     offset = hostbuf.size
@@ -105,7 +105,7 @@ def pin_memory(module, subset="weights", size=None):
         pin.untyped_storage()._comfy_hostbuf = hostbuf
         if torch.cuda.cudart().cudaHostRegister(pin.data_ptr(), size, 1) != 0:
             comfy.model_management.discard_cuda_async_error()
-            comfy.model_management.free_registrations(size, loaded=loaded)
+            comfy.model_management.free_registrations(size)
             if torch.cuda.cudart().cudaHostRegister(pin.data_ptr(), size, 1) != 0:
                 comfy.model_management.discard_cuda_async_error()
                 del pin

comfy_api_nodes/nodes_minimax.py

@@ -467,7 +467,7 @@ def _hailuo03_model_inputs(include_ratio: bool = True, allow_adaptive: bool = Tr
         ),
         IO.Combo.Input(
             "resolution",
-            options=["2K"],
+            options=["768P", "2K"],
             tooltip="Resolution of the output video.",
         ),
     ]
@@ -578,11 +578,12 @@ def define_schema(cls):
             ],
             is_api_node=True,
             price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model.duration"]),
+                depends_on=IO.PriceBadgeDepends(widgets=["model.resolution", "model.duration"]),
                 expr="""
                 (
                   $dur := $lookup(widgets, "model.duration");
-                  {"type": "usd", "usd": $dur * 0.1859}
+                  $rate := $lookup(widgets, "model.resolution") = "768p" ? 0.1287 : 0.1859;
+                  {"type": "usd", "usd": $dur * $rate}
                 )
                 """,
             ),
@@ -660,11 +661,12 @@ def define_schema(cls):
             ],
             is_api_node=True,
             price_badge=IO.PriceBadge(
-                depends_on=IO.PriceBadgeDepends(widgets=["model.duration"]),
+                depends_on=IO.PriceBadgeDepends(widgets=["model.resolution", "model.duration"]),
                 expr="""
                 (
                   $dur := $lookup(widgets, "model.duration");
-                  {"type": "usd", "usd": $dur * 0.1859}
+                  $rate := $lookup(widgets, "model.resolution") = "768p" ? 0.1287 : 0.1859;
+                  {"type": "usd", "usd": $dur * $rate}
                 )
                 """,
             ),
@@ -818,20 +820,21 @@ def define_schema(cls):
             is_api_node=True,
             price_badge=IO.PriceBadge(
                 depends_on=IO.PriceBadgeDepends(
-                    widgets=["model.duration"],
+                    widgets=["model.resolution", "model.duration"],
                     input_groups=["model.reference_images", "model.reference_videos"],
                 ),
                 expr="""
                 (
                   $dur := $lookup(widgets, "model.duration");
+                  $rate := $lookup(widgets, "model.resolution") = "768p" ? 0.1287 : 0.1859;
                   $imgsRaw := $lookup(inputGroups, "model.reference_images");
                   $imgs := $imgsRaw ? $imgsRaw : 0;
                   $vidsRaw := $lookup(inputGroups, "model.reference_videos");
                   $vids := $vidsRaw ? $vidsRaw : 0;
-                  $base := $dur * 0.1859 + ($imgs > 5 ? ($imgs - 5) * 0.0572 : 0);
+                  $base := $dur * $rate + ($imgs > 5 ? ($imgs - 5) * 0.0572 : 0);
                   $vids > 0
-                    ? {"type": "range_usd", "min_usd": $base + $vids * 2 * 0.1859,
-                       "max_usd": $base + 15 * 0.1859, "format": {"approximate": true}}
+                    ? {"type": "range_usd", "min_usd": $base + $vids * 2 * $rate,
+                       "max_usd": $base + 15 * $rate, "format": {"approximate": true}}
                     : {"type": "usd", "usd": $base}
                 )
                 """,

main.py

@@ -58,11 +58,16 @@ def dump_traceback_on_sigint(signum, frame):
 import comfy_aimdo.control
 
 if enables_dynamic_vram():
+    simple_vram_headroom = None if args.reserve_vram is None else int(args.reserve_vram * 1024 ** 3)
     try:
-        comfy_aimdo.control.init(simple_vram_headroom=None if args.reserve_vram is None else int(args.reserve_vram * 1024 ** 3))
+        comfy_aimdo.control.init(simple_vram_headroom=simple_vram_headroom, nvml_pressure=not args.disable_nvml_pressure)
     except TypeError:
-        # comfy-aimdo 0.4.9 protocol.
-        comfy_aimdo.control.init()
+        # comfy-aimdo 0.4.10 protocol.
+        try:
+            comfy_aimdo.control.init(simple_vram_headroom=simple_vram_headroom)
+        except TypeError:
+            # comfy-aimdo 0.4.9 protocol.
+            comfy_aimdo.control.init()
 
 if os.name == "nt":
     os.environ['MIMALLOC_PURGE_DELAY'] = '0'

requirements.txt

@@ -23,7 +23,7 @@ SQLAlchemy>=2.0.0
 filelock
 av>=16.0.0
 comfy-kitchen==0.2.26
-comfy-aimdo==0.4.10
+comfy-aimdo==0.4.11
 requests
 simpleeval>=1.0.0
 blake3