mirror of
https://github.com/LostRuins/koboldcpp.git
synced 2025-09-11 01:24:36 +00:00
configurable resolution limit (#1586)
* refactor image gen configuration screen
* make image size limit configurable
* fix resolution limits and keep dimensions closer to the original ratio
* use 0.0 for the configured default image size limit
This prevents the current default value from being saved into the
config files, in case we later decide to adopt a different value.
* export image model version when loading
* restore model-specific default image size limit
* change the image area restriction to be specified by a square side
* move image resolution limits down to the C++ level
* Revert "export image model version when loading"
This reverts commit fa65b23de3.
* Linting Fixes:
PY:
- Inconsistent var name sd_restrict_square -> sd_restrict_square_var
- GUI swap back to using absolute row numbers for now.
- fstring fix
- size_limit -> side_limit inconsistency
C++:
- roundup_64 standalone function
- refactor sd_fix_resolution variable names for clarity
- move "anti crashing" hard total megapixel limit always to be applied after soft total megapixel limit instead of conditionally only when sd_restrict_square is unset
* allow unsafe resolutions if debugmode is on
---------
Co-authored-by: Concedo <39025047+LostRuins@users.noreply.github.com>
This commit is contained in:
Wagner Bruna
GitHub
parent
f1c9db4174
commit
f6d2d1ce5c
3 changed files with 147 additions and 43 deletions
48
koboldcpp.py
48
koboldcpp.py
|
|
@ -274,6 +274,8 @@ class sd_load_model_inputs(ctypes.Structure):
|
|||
("vae_filename", ctypes.c_char_p),
|
||||
("lora_filename", ctypes.c_char_p),
|
||||
("lora_multiplier", ctypes.c_float),
|
||||
("side_limit", ctypes.c_int),
|
||||
("square_limit", ctypes.c_int),
|
||||
("quiet", ctypes.c_bool),
|
||||
("debugmode", ctypes.c_int)]
|
||||
|
||||
|
|
@ -1539,6 +1541,8 @@ def sd_load_model(model_filename,vae_filename,lora_filename,t5xxl_filename,clipl
|
|||
inputs.t5xxl_filename = t5xxl_filename.encode("UTF-8")
|
||||
inputs.clipl_filename = clipl_filename.encode("UTF-8")
|
||||
inputs.clipg_filename = clipg_filename.encode("UTF-8")
|
||||
inputs.side_limit = args.sdclamped
|
||||
inputs.square_limit = args.sdrestrictsquare
|
||||
inputs = set_backend_props(inputs)
|
||||
ret = handle.sd_load_model(inputs)
|
||||
return ret
|
||||
|
|
@ -1618,27 +1622,11 @@ def sd_generate(genparams):
|
|||
clip_skip = tryparseint(genparams.get("clip_skip", -1),-1)
|
||||
|
||||
#clean vars
|
||||
width = width - (width%64)
|
||||
height = height - (height%64)
|
||||
cfg_scale = (1 if cfg_scale < 1 else (25 if cfg_scale > 25 else cfg_scale))
|
||||
sample_steps = (1 if sample_steps < 1 else (forced_steplimit if sample_steps > forced_steplimit else sample_steps))
|
||||
reslimit = 1024
|
||||
width = (64 if width < 64 else width)
|
||||
height = (64 if height < 64 else height)
|
||||
|
||||
if args.sdclamped:
|
||||
sample_steps = (40 if sample_steps > 40 else sample_steps)
|
||||
reslimit = int(args.sdclamped)
|
||||
reslimit = (512 if reslimit<512 else reslimit)
|
||||
print(f"\nImgGen: Clamped Mode (For Shared Use). Step counts and resolution are clamped to {reslimit}x{reslimit}.")
|
||||
|
||||
biggest = max(width,height)
|
||||
if biggest > reslimit:
|
||||
scaler = biggest / reslimit
|
||||
width = int(width / scaler)
|
||||
height = int(height / scaler)
|
||||
width = width - (width%64)
|
||||
height = height - (height%64)
|
||||
|
||||
inputs = sd_generation_inputs()
|
||||
inputs.prompt = prompt.encode("UTF-8")
|
||||
|
|
@ -4268,6 +4256,7 @@ def show_gui():
|
|||
sd_vaeauto_var = ctk.IntVar(value=0)
|
||||
sd_notile_var = ctk.IntVar(value=0)
|
||||
sd_clamped_var = ctk.StringVar(value="0")
|
||||
sd_restrict_square_var = ctk.StringVar(value="0")
|
||||
sd_threads_var = ctk.StringVar(value=str(default_threads))
|
||||
sd_quant_var = ctk.IntVar(value=0)
|
||||
|
||||
|
|
@ -4965,20 +4954,22 @@ def show_gui():
|
|||
images_tab = tabcontent["Image Gen"]
|
||||
makefileentry(images_tab, "Image Gen. Model (safetensors/gguf):", "Select Image Gen Model File", sd_model_var, 1, width=280, singlecol=True, filetypes=[("*.safetensors *.gguf","*.safetensors *.gguf")], tooltiptxt="Select a .safetensors or .gguf Image Generation model file on disk to be loaded.")
|
||||
makelabelentry(images_tab, "Clamped Mode (Limit Resolution):", sd_clamped_var, 4, 50, padx=290,singleline=True,tooltip="Limit generation steps and resolution settings for shared use.\nSet to 0 to disable, otherwise value is the size limit (min 512px).")
|
||||
makelabelentry(images_tab, "Image Threads:" , sd_threads_var, 6, 50,padx=290,singleline=True,tooltip="How many threads to use during image generation.\nIf left blank, uses same value as threads.")
|
||||
makelabelentry(images_tab, "Restrict Square Size:", sd_restrict_square_var, 6, 50, padx=290,singleline=True,tooltip="Square image size restriction, to protect the server against memory crashes.\nAllows width-height tradeoffs, eg. 640 allows 640x640 and 512x768\nLeave at 0 for the default value: 832 for SD1.5/SD2, 1024 otherwise.")
|
||||
makelabelentry(images_tab, "Image Threads:" , sd_threads_var, 8, 50,padx=290,singleline=True,tooltip="How many threads to use during image generation.\nIf left blank, uses same value as threads.")
|
||||
sd_model_var.trace("w", gui_changed_modelfile)
|
||||
|
||||
makefileentry(images_tab, "Image LoRA (safetensors/gguf):", "Select SD lora file",sd_lora_var, 10, width=280, singlecol=True, filetypes=[("*.safetensors *.gguf", "*.safetensors *.gguf")],tooltiptxt="Select a .safetensors or .gguf SD LoRA model file to be loaded. Should be unquantized!")
|
||||
makelabelentry(images_tab, "Image LoRA Multiplier:" , sd_loramult_var, 12, 50,padx=290,singleline=True,tooltip="What mutiplier value to apply the SD LoRA with.")
|
||||
|
||||
makecheckbox(images_tab, "Compress Weights (Saves Memory)", sd_quant_var, 8,tooltiptxt="Quantizes the SD model weights to save memory. May degrade quality.")
|
||||
makecheckbox(images_tab, "Compress Weights (Saves Memory)", sd_quant_var, 10,tooltiptxt="Quantizes the SD model weights to save memory. May degrade quality.")
|
||||
sd_quant_var.trace("w", changed_gpulayers_estimate)
|
||||
|
||||
makefileentry(images_tab, "T5-XXL File:", "Select Optional T5-XXL model file (SD3 or flux)",sd_t5xxl_var, 14, width=280, singlerow=True, filetypes=[("*.safetensors *.gguf","*.safetensors *.gguf")],tooltiptxt="Select a .safetensors t5xxl file to be loaded.")
|
||||
makefileentry(images_tab, "Clip-L File:", "Select Optional Clip-L model file (SD3 or flux)",sd_clipl_var, 16, width=280, singlerow=True, filetypes=[("*.safetensors *.gguf","*.safetensors *.gguf")],tooltiptxt="Select a .safetensors t5xxl file to be loaded.")
|
||||
makefileentry(images_tab, "Clip-G File:", "Select Optional Clip-G model file (SD3)",sd_clipg_var, 18, width=280, singlerow=True, filetypes=[("*.safetensors *.gguf","*.safetensors *.gguf")],tooltiptxt="Select a .safetensors t5xxl file to be loaded.")
|
||||
makefileentry(images_tab, "Image LoRA (safetensors/gguf):", "Select SD lora file",sd_lora_var, 20, width=280, singlecol=True, filetypes=[("*.safetensors *.gguf", "*.safetensors *.gguf")],tooltiptxt="Select a .safetensors or .gguf SD LoRA model file to be loaded. Should be unquantized!")
|
||||
makelabelentry(images_tab, "Image LoRA Multiplier:" , sd_loramult_var, 22, 50,padx=290,singleline=True,tooltip="What mutiplier value to apply the SD LoRA with.")
|
||||
|
||||
sdvaeitem1,sdvaeitem2,sdvaeitem3 = makefileentry(images_tab, "Image VAE:", "Select Optional SD VAE file",sd_vae_var, 20, width=280, singlerow=True, filetypes=[("*.safetensors *.gguf", "*.safetensors *.gguf")],tooltiptxt="Select a .safetensors or .gguf SD VAE file to be loaded.")
|
||||
|
||||
|
||||
makefileentry(images_tab, "T5-XXL File:", "Select Optional T5-XXL model file (SD3 or flux)",sd_t5xxl_var, 24, width=280, singlerow=True, filetypes=[("*.safetensors *.gguf","*.safetensors *.gguf")],tooltiptxt="Select a .safetensors t5xxl file to be loaded.")
|
||||
makefileentry(images_tab, "Clip-L File:", "Select Optional Clip-L model file (SD3 or flux)",sd_clipl_var, 26, width=280, singlerow=True, filetypes=[("*.safetensors *.gguf","*.safetensors *.gguf")],tooltiptxt="Select a .safetensors t5xxl file to be loaded.")
|
||||
makefileentry(images_tab, "Clip-G File:", "Select Optional Clip-G model file (SD3)",sd_clipg_var, 28, width=280, singlerow=True, filetypes=[("*.safetensors *.gguf","*.safetensors *.gguf")],tooltiptxt="Select a .safetensors t5xxl file to be loaded.")
|
||||
|
||||
sdvaeitem1,sdvaeitem2,sdvaeitem3 = makefileentry(images_tab, "Image VAE:", "Select Optional SD VAE file",sd_vae_var, 30, width=280, singlerow=True, filetypes=[("*.safetensors *.gguf", "*.safetensors *.gguf")],tooltiptxt="Select a .safetensors or .gguf SD VAE file to be loaded.")
|
||||
def toggletaesd(a,b,c):
|
||||
if sd_vaeauto_var.get()==1:
|
||||
sdvaeitem1.grid_remove()
|
||||
|
|
@ -4989,7 +4980,7 @@ def show_gui():
|
|||
sdvaeitem1.grid()
|
||||
sdvaeitem2.grid()
|
||||
sdvaeitem3.grid()
|
||||
makecheckbox(images_tab, "Use TAE SD (AutoFix Broken VAE)", sd_vaeauto_var, 22,command=toggletaesd,tooltiptxt="Replace VAE with TAESD. May fix bad VAE.")
|
||||
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, 24,tooltiptxt="Disables VAE tiling, may not work for large images.")
|
||||
|
||||
# audio tab
|
||||
|
|
@ -5222,6 +5213,7 @@ def show_gui():
|
|||
|
||||
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.sdrestrictsquare = (0 if int(sd_restrict_square_var.get())<=0 else int(sd_restrict_square_var.get()))
|
||||
args.sdnotile = (True if sd_notile_var.get()==1 else False)
|
||||
if sd_vaeauto_var.get()==1:
|
||||
args.sdvaeauto = True
|
||||
|
|
@ -5432,6 +5424,7 @@ def show_gui():
|
|||
|
||||
sd_model_var.set(dict["sdmodel"] if ("sdmodel" in dict and dict["sdmodel"]) else "")
|
||||
sd_clamped_var.set(int(dict["sdclamped"]) if ("sdclamped" in dict and dict["sdclamped"]) else 0)
|
||||
sd_restrict_square_var.set(int(dict["sdrestrictsquare"]) if ("sdrestrictsquare" in dict and dict["sdrestrictsquare"]) else 0)
|
||||
sd_threads_var.set(str(dict["sdthreads"]) if ("sdthreads" in dict and dict["sdthreads"]) else str(default_threads))
|
||||
sd_quant_var.set(1 if ("sdquant" in dict and dict["sdquant"]) else 0)
|
||||
sd_vae_var.set(dict["sdvae"] if ("sdvae" in dict and dict["sdvae"]) else "")
|
||||
|
|
@ -7168,6 +7161,7 @@ if __name__ == '__main__':
|
|||
sdparsergroup.add_argument("--sdmodel", metavar=('[filename]'), help="Specify an image generation safetensors or gguf model to enable image generation.", default="")
|
||||
sdparsergroup.add_argument("--sdthreads", metavar=('[threads]'), help="Use a different number of threads for image generation if specified. Otherwise, has the same value as --threads.", type=int, default=0)
|
||||
sdparsergroup.add_argument("--sdclamped", metavar=('[maxres]'), help="If specified, limit generation steps and resolution settings for shared use. Accepts an extra optional parameter that indicates maximum resolution (eg. 768 clamps to 768x768, min 512px, disabled if 0).", nargs='?', const=512, type=int, default=0)
|
||||
sdparsergroup.add_argument("--sdrestrictsquare", metavar=('[maxres]'), help="If specified, restrict square image sides to this value, in pixels, to avoid server crashes related to excessive memory usage. Similar to --sdclamped, but allows trade-offs between width and height (e.g. 640 would allow 640x640, 512x768 and 768x512 images). If 0 or unspecified, use a model-specific safe value: 832 for SD1.5/SD2, 1024 otherwise. Total resolution cannot exceed 1MP.", type=int, default=0)
|
||||
sdparsergroup.add_argument("--sdt5xxl", metavar=('[filename]'), help="Specify a T5-XXL safetensors model for use in SD3 or Flux. Leave blank if prebaked or unused.", default="")
|
||||
sdparsergroup.add_argument("--sdclipl", metavar=('[filename]'), help="Specify a Clip-L safetensors model for use in SD3 or Flux. Leave blank if prebaked or unused.", default="")
|
||||
sdparsergroup.add_argument("--sdclipg", metavar=('[filename]'), help="Specify a Clip-G safetensors model for use in SD3. Leave blank if prebaked or unused.", default="")
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue