Automatically enable lowvram mode if vram is less than 4GB.

Use: --normalvram to disable it.

comfy/model_management.py

@@ -13,16 +13,22 @@ total_vram_available_mb = -1
 import sys
 
 set_vram_to = NORMAL_VRAM
+
+try:
+    import torch
+    total_vram = torch.cuda.mem_get_info(torch.cuda.current_device())[1] / (1024 * 1024)
+    if total_vram <= 4096 and not "--normalvram" in sys.argv:
+        print("Trying to enable lowvram mode because your GPU seems to have 4GB or less. If you don't want this use: --normalvram")
+        set_vram_to = LOW_VRAM
+except:
+    pass
+
 if "--lowvram" in sys.argv:
     set_vram_to = LOW_VRAM
 if "--novram" in sys.argv:
     set_vram_to = NO_VRAM
 
-try:
-    import torch
-    total_vram = torch.cuda.mem_get_info(torch.cuda.current_device())[1] / (1024 * 1024)
-except:
-    pass
+
 
 if set_vram_to != NORMAL_VRAM:
     try:

main.py

@@ -14,6 +14,7 @@ if __name__ == "__main__":
         print("\t--dont-upcast-attention\t\tDisable upcasting of attention \n\t\t\t\t\tcan boost speed but increase the chances of black images.\n")
         print("\t--use-split-cross-attention\tUse the split cross attention optimization instead of the sub-quadratic one.\n\t\t\t\t\tIgnored when xformers is used.")
         print()
+        print("\t--normalvram\t\t\tUsed to force normal vram use if lowvram gets automatically enabled.")
         print("\t--lowvram\t\t\tSplit the unet in parts to use less vram.")
         print("\t--novram\t\t\tWhen lowvram isn't enough.")
         print()