Replace the notile option with a configurable threshold

This allows selecting a lower threshold value, reducing the
peak memory usage.

The legacy sdnotile parameter gets automatically converted to
the new parameter, if it's the only one supplied.

Author: wbruna

koboldcpp.py

@@ -1540,7 +1540,7 @@ def sd_load_model(model_filename,vae_filename,lora_filename,t5xxl_filename,clipl
     inputs.threads = thds
     inputs.quant = quant
     inputs.taesd = True if args.sdvaeauto else False
-    inputs.tiled_vae_threshold = -1 if args.sdnotile else 0
+    inputs.tiled_vae_threshold = args.sdtiledvae
     inputs.vae_filename = vae_filename.encode("UTF-8")
     inputs.lora_filename = lora_filename.encode("UTF-8")
     inputs.lora_multiplier = args.sdloramult
@@ -4289,7 +4289,7 @@ def hide_tooltip(event):
     sd_clipl_var = ctk.StringVar()
     sd_clipg_var = ctk.StringVar()
     sd_vaeauto_var = ctk.IntVar(value=0)
-    sd_notile_var = ctk.IntVar(value=0)
+    sd_tiled_vae_var = ctk.StringVar(value="")
     sd_clamped_var = ctk.StringVar(value="0")
     sd_clamped_soft_var = ctk.StringVar(value="0")
     sd_threads_var = ctk.StringVar(value=str(default_threads))
@@ -5020,7 +5020,7 @@ def toggletaesd(a,b,c):
                 sdvaeitem2.grid()
                 sdvaeitem3.grid()
     makecheckbox(images_tab, "Use TAE SD (AutoFix Broken VAE)", sd_vaeauto_var, 32,command=toggletaesd,tooltiptxt="Replace VAE with TAESD. May fix bad VAE.")
-    makecheckbox(images_tab, "No VAE Tiling", sd_notile_var, 34,tooltiptxt="Disables VAE tiling, may not work for large images.")
+    makelabelentry(images_tab, "VAE Tiling Above:", sd_tiled_vae_var, 34, 50, padx=190,singleline=True,tooltip="Enable VAE Tiling for images above this size, to save memory.\n-1 disables tiling for all images. Leave at 0 for the default limit (768).")
 
     # audio tab
     audio_tab = tabcontent["Audio"]
@@ -5253,7 +5253,7 @@ def export_vars():
         args.sdthreads = (0 if sd_threads_var.get()=="" else int(sd_threads_var.get()))
         args.sdclamped = (0 if int(sd_clamped_var.get())<=0 else int(sd_clamped_var.get()))
         args.sdclampedsoft = (0 if int(sd_clamped_soft_var.get())<=0 else int(sd_clamped_soft_var.get()))
-        args.sdnotile = (True if sd_notile_var.get()==1 else False)
+        args.sdtiledvae = (0 if sd_tiled_vae_var.get()=="" else int(sd_tiled_vae_var.get()))
         if sd_vaeauto_var.get()==1:
             args.sdvaeauto = True
             args.sdvae = ""
@@ -5472,7 +5472,8 @@ def import_vars(dict):
         sd_clipl_var.set(dict["sdclipl"] if ("sdclipl" in dict and dict["sdclipl"]) else "")
         sd_clipg_var.set(dict["sdclipg"] if ("sdclipg" in dict and dict["sdclipg"]) else "")
         sd_vaeauto_var.set(1 if ("sdvaeauto" in dict and dict["sdvaeauto"]) else 0)
-        sd_notile_var.set(1 if ("sdnotile" in dict and dict["sdnotile"]) else 0)
+        sd_tiled_vae_var.set(int(dict["sdtiledvae"]) if ("sdtiledvae" in dict and dict["sdtiledvae"]) else 0)
+
         sd_lora_var.set(dict["sdlora"] if ("sdlora" in dict and dict["sdlora"]) else "")
         sd_loramult_var.set(str(dict["sdloramult"]) if ("sdloramult" in dict and dict["sdloramult"]) else "1.0")
 
@@ -5840,6 +5841,8 @@ def convert_invalid_args(args):
             dict["model_param"] = model_value
         elif isinstance(model_value, list) and model_value:  # Non-empty list
             dict["model_param"] = model_value[0]  # Take the first file in the list
+    if "sdnotile" in dict and ("sdtiledvae" not in dict or dict["sdtiledvae"] == 0):
+        dict["sdtiledvae"] = (-1 if dict["sdnotile"] else 0) # convert legacy option
     return args
 
 def setuptunnel(global_memory, has_sd):
@@ -7242,8 +7245,8 @@ def range_checker(arg: str):
     sdparsergrouplora.add_argument("--sdquant", help="If specified, loads the model quantized to save memory.", action='store_true')
     sdparsergrouplora.add_argument("--sdlora", metavar=('[filename]'), help="Specify an image generation LORA safetensors model to be applied.", default="")
     sdparsergroup.add_argument("--sdloramult", metavar=('[amount]'), help="Multiplier for the image LORA model to be applied.", type=float, default=1.0)
-    sdparsergroup.add_argument("--sdnotile", help="Disables VAE tiling, may not work for large images.", action='store_true')
-
+    sdparsergroup.add_argument("--sdtiledvae", metavar=('[maxres]'), help="Adjust the automatic VAE tiling trigger for images above this size. 1 will always use tiling, -1 disables it for all images; 0 or unspecified uses the default 768", type=int, default=0)
+    sdparsergroup.add_argument("--sdnotile", help=argparse.SUPPRESS, action='store_true') # legacy option, see sdtiledvae
     whisperparsergroup = parser.add_argument_group('Whisper Transcription Commands')
     whisperparsergroup.add_argument("--whispermodel", metavar=('[filename]'), help="Specify a Whisper .bin model to enable Speech-To-Text transcription.", default="")