KSampler (Efficient)¶
Documentation¶
- Class name:
KSampler (Efficient)
- Category:
Efficiency Nodes/Sampling
- Output node:
True
The KSampler (Efficient) node is designed to efficiently sample latent images using a variety of sampling strategies, including regular and advanced methods. It leverages the Comfy KSampler nodes to generate or refine latent images based on specified conditions, aiming to optimize the sampling process for speed and resource usage.
Input types¶
Required¶
model
- Specifies the model to be used for sampling. It is crucial for determining the sampling behavior and the quality of the generated images.
- Comfy dtype:
MODEL
- Python dtype:
torch.nn.Module
seed
- Determines the random seed for sampling, ensuring reproducibility of results. It affects the randomness of the generated latent images.
- Comfy dtype:
INT
- Python dtype:
int
steps
- Defines the number of steps to perform during sampling. It influences the detail and quality of the generated images.
- Comfy dtype:
INT
- Python dtype:
int
cfg
- Controls the conditioning-free guidance (CFG) scale, affecting the adherence to the specified conditions.
- Comfy dtype:
FLOAT
- Python dtype:
float
sampler_name
- Identifies the specific sampler algorithm to use, impacting the sampling method and results.
- Comfy dtype:
COMBO[STRING]
- Python dtype:
str
scheduler
- Specifies the scheduler for controlling the sampling process, affecting the progression of image generation.
- Comfy dtype:
COMBO[STRING]
- Python dtype:
str
positive
- The positive conditioning text to guide the image generation towards desired attributes.
- Comfy dtype:
CONDITIONING
- Python dtype:
str
negative
- The negative conditioning text to steer the image generation away from undesired attributes.
- Comfy dtype:
CONDITIONING
- Python dtype:
str
latent_image
- The initial latent image to refine or generate from, if provided. It influences the starting point of the sampling process.
- Comfy dtype:
LATENT
- Python dtype:
torch.Tensor
denoise
- Adjusts the level of denoising applied during sampling, affecting the clarity and quality of the generated images.
- Comfy dtype:
FLOAT
- Python dtype:
float
preview_method
- unknown
- Comfy dtype:
COMBO[STRING]
- Python dtype:
unknown
vae_decode
- unknown
- Comfy dtype:
COMBO[STRING]
- Python dtype:
unknown
Optional¶
optional_vae
- unknown
- Comfy dtype:
VAE
- Python dtype:
unknown
script
- unknown
- Comfy dtype:
SCRIPT
- Python dtype:
unknown
Output types¶
MODEL
- Comfy dtype:
MODEL
- unknown
- Python dtype:
unknown
- Comfy dtype:
CONDITIONING+
- Comfy dtype:
CONDITIONING
- Positive conditioning information that has been applied or generated during the sampling process.
- Python dtype:
str
- Comfy dtype:
CONDITIONING-
- Comfy dtype:
CONDITIONING
- Negative conditioning information that has been applied or generated during the sampling process.
- Python dtype:
str
- Comfy dtype:
LATENT
- Comfy dtype:
LATENT
- The sampled or refined latent image, representing the result of the sampling process.
- Python dtype:
torch.Tensor
- Comfy dtype:
VAE
- Comfy dtype:
VAE
- VAE-related output, potentially including encoded or decoded information as part of the sampling process.
- Python dtype:
torch.Tensor
- Comfy dtype:
IMAGE
- Comfy dtype:
IMAGE
- The final image output, representing the visual result of the sampling and conditioning process.
- Python dtype:
torch.Tensor
- Comfy dtype:
Usage tips¶
- Infra type:
GPU
- Common nodes:
Source code¶
class TSC_KSampler:
empty_image = pil2tensor(Image.new('RGBA', (1, 1), (0, 0, 0, 0)))
@classmethod
def INPUT_TYPES(cls):
return {"required":
{"model": ("MODEL",),
"seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
"steps": ("INT", {"default": 20, "min": 1, "max": 10000}),
"cfg": ("FLOAT", {"default": 7.0, "min": 0.0, "max": 100.0}),
"sampler_name": (comfy.samplers.KSampler.SAMPLERS,),
"scheduler": (SCHEDULERS,),
"positive": ("CONDITIONING",),
"negative": ("CONDITIONING",),
"latent_image": ("LATENT",),
"denoise": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
"preview_method": (["auto", "latent2rgb", "taesd", "vae_decoded_only", "none"],),
"vae_decode": (["true", "true (tiled)", "false"],),
},
"optional": { "optional_vae": ("VAE",),
"script": ("SCRIPT",),},
"hidden": {"prompt": "PROMPT", "extra_pnginfo": "EXTRA_PNGINFO", "my_unique_id": "UNIQUE_ID",},
}
RETURN_TYPES = ("MODEL", "CONDITIONING", "CONDITIONING", "LATENT", "VAE", "IMAGE", )
RETURN_NAMES = ("MODEL", "CONDITIONING+", "CONDITIONING-", "LATENT", "VAE", "IMAGE", )
OUTPUT_NODE = True
FUNCTION = "sample"
CATEGORY = "Efficiency Nodes/Sampling"
def sample(self, model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image,
preview_method, vae_decode, denoise=1.0, prompt=None, extra_pnginfo=None, my_unique_id=None,
optional_vae=(None,), script=None, add_noise=None, start_at_step=None, end_at_step=None,
return_with_leftover_noise=None, sampler_type="regular"):
# Rename the vae variable
vae = optional_vae
# If vae is not connected, disable vae decoding
if vae == (None,) and vae_decode != "false":
print(f"{warning('KSampler(Efficient) Warning:')} No vae input detected, proceeding as if vae_decode was false.\n")
vae_decode = "false"
#---------------------------------------------------------------------------------------------------------------
# Unpack SDXL Tuple embedded in the 'model' channel
if sampler_type == "sdxl":
sdxl_tuple = model
model, _, positive, negative, refiner_model, _, refiner_positive, refiner_negative = sdxl_tuple
else:
refiner_model = refiner_positive = refiner_negative = None
#---------------------------------------------------------------------------------------------------------------
def keys_exist_in_script(*keys):
return any(key in script for key in keys) if script else False
#---------------------------------------------------------------------------------------------------------------
def vae_decode_latent(vae, samples, vae_decode):
return VAEDecodeTiled().decode(vae,samples,320)[0] if "tiled" in vae_decode else VAEDecode().decode(vae,samples)[0]
def vae_encode_image(vae, pixels, vae_decode):
return VAEEncodeTiled().encode(vae,pixels,320)[0] if "tiled" in vae_decode else VAEEncode().encode(vae,pixels)[0]
# ---------------------------------------------------------------------------------------------------------------
def process_latent_image(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image,
denoise, sampler_type, add_noise, start_at_step, end_at_step, return_with_leftover_noise,
refiner_model, refiner_positive, refiner_negative, vae, vae_decode, preview_method):
# Store originals
original_calculation = comfy.samplers.calculate_sigmas
original_KSampler_SCHEDULERS = comfy.samplers.KSampler.SCHEDULERS
previous_preview_method = global_preview_method()
original_prepare_noise = comfy.sample.prepare_noise
original_KSampler = comfy.samplers.KSampler
original_model_str = str(model)
# monkey patch the sample function
def calculate_sigmas(model_sampling, scheduler_name: str, steps):
if scheduler_name.startswith("AYS"):
return AlignYourStepsScheduler().get_sigmas(scheduler_name.split(" ")[1], steps, denoise=1.0)[0]
return original_calculation(model_sampling, scheduler_name, steps)
comfy.samplers.KSampler.SCHEDULERS = SCHEDULERS
comfy.samplers.calculate_sigmas = calculate_sigmas
# Initialize output variables
samples = images = gifs = preview = cnet_imgs = None
try:
# Change the global preview method (temporarily)
set_preview_method(preview_method)
# ------------------------------------------------------------------------------------------------------
# Check if "noise" exists in the script before main sampling has taken place
if keys_exist_in_script("noise"):
rng_source, cfg_denoiser, add_seed_noise, m_seed, m_weight = script["noise"]
smZ_rng_source.rng_rand_source(rng_source) # this function monkey patches comfy.sample.prepare_noise
if cfg_denoiser:
comfy.samplers.KSampler = smZ_cfg_denoiser.SDKSampler
if add_seed_noise:
comfy.sample.prepare_noise = cg_mixed_seed_noise.get_mixed_noise_function(comfy.sample.prepare_noise, m_seed, m_weight)
else:
m_seed = m_weight = None
else:
rng_source = cfg_denoiser = add_seed_noise = m_seed = m_weight = None
# ------------------------------------------------------------------------------------------------------
# Check if "anim" exists in the script before main sampling has taken place
if keys_exist_in_script("anim"):
if preview_method != "none":
set_preview_method("none") # disable preview method
print(f"{warning('KSampler(Efficient) Warning:')} Live preview disabled for animatediff generations.")
motion_model, beta_schedule, context_options, frame_rate, loop_count, format, pingpong, save_image = script["anim"]
model = AnimateDiffLoaderWithContext().load_mm_and_inject_params(model, motion_model, beta_schedule, context_options)[0]
# ------------------------------------------------------------------------------------------------------
# Store run parameters as strings. Load previous stored samples if all parameters match.
latent_image_hash = tensor_to_hash(latent_image["samples"])
positive_hash = tensor_to_hash(positive[0][0])
negative_hash = tensor_to_hash(negative[0][0])
refiner_positive_hash = tensor_to_hash(refiner_positive[0][0]) if refiner_positive is not None else None
refiner_negative_hash = tensor_to_hash(refiner_negative[0][0]) if refiner_negative is not None else None
# Include motion_model, beta_schedule, and context_options as unique identifiers if they exist.
model_identifier = [original_model_str, motion_model, beta_schedule, context_options] if keys_exist_in_script("anim")\
else [original_model_str]
parameters = [model_identifier] + [seed, steps, cfg, sampler_name, scheduler, positive_hash, negative_hash,
latent_image_hash, denoise, sampler_type, add_noise, start_at_step,
end_at_step, return_with_leftover_noise, refiner_model, refiner_positive_hash,
refiner_negative_hash, rng_source, cfg_denoiser, add_seed_noise, m_seed, m_weight]
# Convert all elements in parameters to strings, except for the hash variable checks
parameters = [str(item) if not isinstance(item, type(latent_image_hash)) else item for item in parameters]
# Load previous latent if all parameters match, else returns 'None'
samples = load_ksampler_results("latent", my_unique_id, parameters)
if samples is None: # clear stored images
store_ksampler_results("image", my_unique_id, None)
store_ksampler_results("cnet_img", my_unique_id, None)
if samples is not None: # do not re-sample
images = load_ksampler_results("image", my_unique_id)
cnet_imgs = True # "True" will denote that it can be loaded provided the preprocessor matches
# Sample the latent_image(s) using the Comfy KSampler nodes
elif sampler_type == "regular":
samples = KSampler().sample(model, seed, steps, cfg, sampler_name, scheduler, positive, negative,
latent_image, denoise=denoise)[0] if denoise>0 else latent_image
elif sampler_type == "advanced":
samples = KSamplerAdvanced().sample(model, add_noise, seed, steps, cfg, sampler_name, scheduler,
positive, negative, latent_image, start_at_step, end_at_step,
return_with_leftover_noise, denoise=1.0)[0]
elif sampler_type == "sdxl":
# Disable refiner if refine_at_step is -1
if end_at_step == -1:
end_at_step = steps
# Perform base model sampling
add_noise = return_with_leftover_noise = "enable"
samples = KSamplerAdvanced().sample(model, add_noise, seed, steps, cfg, sampler_name, scheduler,
positive, negative, latent_image, start_at_step, end_at_step,
return_with_leftover_noise, denoise=1.0)[0]
# Perform refiner model sampling
if refiner_model and end_at_step < steps:
add_noise = return_with_leftover_noise = "disable"
samples = KSamplerAdvanced().sample(refiner_model, add_noise, seed, steps, cfg + REFINER_CFG_OFFSET,
sampler_name, scheduler, refiner_positive, refiner_negative,
samples, end_at_step, steps,
return_with_leftover_noise, denoise=1.0)[0]
# Cache the first pass samples in the 'last_helds' dictionary "latent" if not xyplot
if not any(keys_exist_in_script(key) for key in ["xyplot"]):
store_ksampler_results("latent", my_unique_id, samples, parameters)
# ------------------------------------------------------------------------------------------------------
# Check if "hiresfix" exists in the script after main sampling has taken place
if keys_exist_in_script("hiresfix"):
# Unpack the tuple from the script's "hiresfix" key
upscale_type, latent_upscaler, upscale_by, use_same_seed, hires_seed, hires_steps, hires_denoise,\
iterations, hires_control_net, hires_cnet_strength, preprocessor, preprocessor_imgs, \
latent_upscale_function, latent_upscale_model, pixel_upscale_model = script["hiresfix"]
# Define hires_seed
hires_seed = seed if use_same_seed else hires_seed
# Define latent_upscale_model
if latent_upscale_model is None:
latent_upscale_model = model
elif keys_exist_in_script("anim"):
latent_upscale_model = \
AnimateDiffLoaderWithContext().load_mm_and_inject_params(latent_upscale_model, motion_model,
beta_schedule, context_options)[0]
# Generate Preprocessor images and Apply Control Net
if hires_control_net is not None:
# Attempt to load previous "cnet_imgs" if previous images were loaded and preprocessor is same
if cnet_imgs is True:
cnet_imgs = load_ksampler_results("cnet_img", my_unique_id, [preprocessor])
# If cnet_imgs is None, generate new ones
if cnet_imgs is None:
if images is None:
images = vae_decode_latent(vae, samples, vae_decode)
store_ksampler_results("image", my_unique_id, images)
cnet_imgs = AIO_Preprocessor().execute(preprocessor, images)[0]
store_ksampler_results("cnet_img", my_unique_id, cnet_imgs, [preprocessor])
positive = ControlNetApply().apply_controlnet(positive, hires_control_net, cnet_imgs, hires_cnet_strength)[0]
# Iterate for the given number of iterations
if upscale_type == "latent":
for _ in range(iterations):
upscaled_latent_image = latent_upscale_function().upscale(samples, latent_upscaler, upscale_by)[0]
samples = KSampler().sample(latent_upscale_model, hires_seed, hires_steps, cfg, sampler_name, scheduler,
positive, negative, upscaled_latent_image, denoise=hires_denoise)[0]
images = None # set to None when samples is updated
elif upscale_type == "pixel":
if images is None:
images = vae_decode_latent(vae, samples, vae_decode)
store_ksampler_results("image", my_unique_id, images)
images = ImageUpscaleWithModel().upscale(pixel_upscale_model, images)[0]
images = ImageScaleBy().upscale(images, "nearest-exact", upscale_by/4)[0]
elif upscale_type == "both":
for _ in range(iterations):
if images is None:
images = vae_decode_latent(vae, samples, vae_decode)
store_ksampler_results("image", my_unique_id, images)
images = ImageUpscaleWithModel().upscale(pixel_upscale_model, images)[0]
images = ImageScaleBy().upscale(images, "nearest-exact", upscale_by/4)[0]
samples = vae_encode_image(vae, images, vae_decode)
upscaled_latent_image = latent_upscale_function().upscale(samples, latent_upscaler, 1)[0]
samples = KSampler().sample(latent_upscale_model, hires_seed, hires_steps, cfg, sampler_name, scheduler,
positive, negative, upscaled_latent_image, denoise=hires_denoise)[0]
images = None # set to None when samples is updated
# ------------------------------------------------------------------------------------------------------
# Check if "tile" exists in the script after main sampling has taken place
if keys_exist_in_script("tile"):
# Unpack the tuple from the script's "tile" key
upscale_by, tile_size, tiling_strategy, tiling_steps, tile_seed, tiled_denoise,\
tile_controlnet, strength = script["tile"]
# Decode image, store if first decode
if images is None:
images = vae_decode_latent(vae, samples, vae_decode)
if not any(keys_exist_in_script(key) for key in ["xyplot", "hiresfix"]):
store_ksampler_results("image", my_unique_id, images)
# Upscale image
upscaled_image = ImageScaleBy().upscale(images, "nearest-exact", upscale_by)[0]
upscaled_latent = vae_encode_image(vae, upscaled_image, vae_decode)
# If using Control Net, Apply Control Net using upscaled_image and loaded control_net
if tile_controlnet is not None:
positive = ControlNetApply().apply_controlnet(positive, tile_controlnet, upscaled_image, 1)[0]
# Sample latent
TSampler = bnk_tiled_samplers.TiledKSampler
samples = TSampler().sample(model, tile_seed, tile_size, tile_size, tiling_strategy, tiling_steps, cfg,
sampler_name, scheduler, positive, negative, upscaled_latent,
denoise=tiled_denoise)[0]
images = None # set to None when samples is updated
# ------------------------------------------------------------------------------------------------------
# Check if "anim" exists in the script after the main sampling has taken place
if keys_exist_in_script("anim"):
if images is None:
images = vae_decode_latent(vae, samples, vae_decode)
if not any(keys_exist_in_script(key) for key in ["xyplot", "hiresfix", "tile"]):
store_ksampler_results("image", my_unique_id, images)
gifs = AnimateDiffCombine().generate_gif(images, frame_rate, loop_count, format=format,
pingpong=pingpong, save_image=save_image, prompt=prompt, extra_pnginfo=extra_pnginfo)["ui"]["gifs"]
# ------------------------------------------------------------------------------------------------------
# Decode image if not yet decoded
if "true" in vae_decode:
if images is None:
images = vae_decode_latent(vae, samples, vae_decode)
# Store decoded image as base image of no script is detected
if all(not keys_exist_in_script(key) for key in ["xyplot", "hiresfix", "tile", "anim"]):
store_ksampler_results("image", my_unique_id, images)
# Append Control Net Images (if exist)
if cnet_imgs is not None and not True:
if preprocessor_imgs and upscale_type == "latent":
if keys_exist_in_script("xyplot"):
print(
f"{warning('HighRes-Fix Warning:')} Preprocessor images auto-disabled when XY Plotting.")
else:
# Resize cnet_imgs if necessary and stack
if images.shape[1:3] != cnet_imgs.shape[1:3]: # comparing height and width
cnet_imgs = quick_resize(cnet_imgs, images.shape)
images = torch.cat([images, cnet_imgs], dim=0)
# Define preview images
if keys_exist_in_script("anim"):
preview = {"gifs": gifs, "images": list()}
elif preview_method == "none" or (preview_method == "vae_decoded_only" and vae_decode == "false"):
preview = {"images": list()}
elif images is not None:
preview = PreviewImage().save_images(images, prompt=prompt, extra_pnginfo=extra_pnginfo)["ui"]
# Define a dummy output image
if images is None and vae_decode == "false":
images = TSC_KSampler.empty_image
finally:
# Restore global changes
set_preview_method(previous_preview_method)
comfy.samplers.KSampler = original_KSampler
comfy.sample.prepare_noise = original_prepare_noise
comfy.samplers.calculate_sigmas = original_calculation
comfy.samplers.KSampler.SCHEDULERS = original_KSampler_SCHEDULERS
return samples, images, gifs, preview
# ---------------------------------------------------------------------------------------------------------------
# Clean globally stored objects of non-existant nodes
globals_cleanup(prompt)
# ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
# If not XY Plotting
if not keys_exist_in_script("xyplot"):
# Process latent image
samples, images, gifs, preview = process_latent_image(model, seed, steps, cfg, sampler_name, scheduler,
positive, negative, latent_image, denoise, sampler_type, add_noise,
start_at_step, end_at_step, return_with_leftover_noise, refiner_model,
refiner_positive, refiner_negative, vae, vae_decode, preview_method)
if sampler_type == "sdxl":
result = (sdxl_tuple, samples, vae, images,)
else:
result = (model, positive, negative, samples, vae, images,)
if preview is None:
return {"result": result}
else:
return {"ui": preview, "result": result}
# ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
# If XY Plot
elif keys_exist_in_script("xyplot"):
# If no vae connected, throw errors
if vae == (None,):
print(f"{error('KSampler(Efficient) Error:')} VAE input must be connected in order to use the XY Plot script.")
return {"ui": {"images": list()},
"result": (model, positive, negative, latent_image, vae, TSC_KSampler.empty_image,)}
# If vae_decode is not set to true, print message that changing it to true
if "true" not in vae_decode:
print(f"{warning('KSampler(Efficient) Warning:')} VAE decoding must be set to \'true\'"
" for the XY Plot script, proceeding as if \'true\'.\n")
#___________________________________________________________________________________________________________
# Initialize, unpack, and clean variables for the XY Plot script
vae_name = None
ckpt_name = None
clip = None
clip_skip = None
refiner_name = None
refiner_clip = None
refiner_clip_skip = None
positive_prompt = None
negative_prompt = None
ascore = None
empty_latent_width = None
empty_latent_height = None
lora_stack = None
cnet_stack = None
# Split the 'samples' tensor
samples_tensors = torch.split(latent_image['samples'], 1, dim=0)
# Check if 'noise_mask' exists and split if it does
if 'noise_mask' in latent_image:
noise_mask_tensors = torch.split(latent_image['noise_mask'], 1, dim=0)
latent_tensors = [{'samples': img, 'noise_mask': mask} for img, mask in
zip(samples_tensors, noise_mask_tensors)]
else:
latent_tensors = [{'samples': img} for img in samples_tensors]
# Set latent only to the first of the batch
latent_image = latent_tensors[0]
# Unpack script Tuple (X_type, X_value, Y_type, Y_value, grid_spacing, Y_label_orientation, dependencies)
X_type, X_value, Y_type, Y_value, grid_spacing, Y_label_orientation, cache_models, xyplot_as_output_image,\
xyplot_id, dependencies = script["xyplot"]
#_______________________________________________________________________________________________________
# The below section is used to check wether the XY_type is allowed for the Ksampler instance being used.
# If not the correct type, this section will abort the xy plot script.
samplers = {
"regular": {
"disallowed": ["AddNoise", "ReturnNoise", "StartStep", "EndStep", "RefineStep",
"Refiner", "Refiner On/Off", "AScore+", "AScore-"],
"name": "KSampler (Efficient)"
},
"advanced": {
"disallowed": ["RefineStep", "Denoise", "RefineStep", "Refiner", "Refiner On/Off",
"AScore+", "AScore-"],
"name": "KSampler Adv. (Efficient)"
},
"sdxl": {
"disallowed": ["AddNoise", "EndStep", "Denoise"],
"name": "KSampler SDXL (Eff.)"
}
}
# Define disallowed XY_types for each ksampler type
def get_ksampler_details(sampler_type):
return samplers.get(sampler_type, {"disallowed": [], "name": ""})
def suggest_ksampler(X_type, Y_type, current_sampler):
for sampler, details in samplers.items():
if sampler != current_sampler and X_type not in details["disallowed"] and Y_type not in details["disallowed"]:
return details["name"]
return "a different KSampler"
# In your main function or code segment:
details = get_ksampler_details(sampler_type)
disallowed_XY_types = details["disallowed"]
ksampler_name = details["name"]
if X_type in disallowed_XY_types or Y_type in disallowed_XY_types:
error_prefix = f"{error(f'{ksampler_name} Error:')}"
failed_type = []
if X_type in disallowed_XY_types:
failed_type.append(f"X_type: '{X_type}'")
if Y_type in disallowed_XY_types:
failed_type.append(f"Y_type: '{Y_type}'")
suggested_ksampler = suggest_ksampler(X_type, Y_type, sampler_type)
print(f"{error_prefix} Invalid value for {' and '.join(failed_type)}. "
f"Use {suggested_ksampler} for this XY Plot type."
f"\nDisallowed XY_types for this KSampler are: {', '.join(disallowed_XY_types)}.")
return {"ui": {"images": list()},
"result": (model, positive, negative, latent_image, vae, TSC_KSampler.empty_image,)}
#_______________________________________________________________________________________________________
# Unpack Effficient Loader dependencies
if dependencies is not None:
vae_name, ckpt_name, clip, clip_skip, refiner_name, refiner_clip, refiner_clip_skip,\
positive_prompt, negative_prompt, token_normalization, weight_interpretation, ascore,\
empty_latent_width, empty_latent_height, lora_stack, cnet_stack = dependencies
#_______________________________________________________________________________________________________
# Printout XY Plot values to be processed
def process_xy_for_print(value, replacement, type_):
if type_ == "Seeds++ Batch" and isinstance(value, list):
return [v + seed for v in value] # Add seed to every entry in the list
elif type_ == "Scheduler" and isinstance(value, tuple):
return value[0] # Return only the first entry of the tuple
elif type_ == "VAE" and isinstance(value, list):
# For each string in the list, extract the filename from the path
return [os.path.basename(v) for v in value]
elif (type_ == "Checkpoint" or type_ == "Refiner") and isinstance(value, list):
# For each tuple in the list, return only the first value if the second or third value is None
return [(os.path.basename(v[0]),) + v[1:] if v[1] is None or v[2] is None
else (os.path.basename(v[0]), v[1]) if v[2] is None
else (os.path.basename(v[0]),) + v[1:] for v in value]
elif type_ == "LoRA" and isinstance(value, list):
# Return only the first Tuple of each inner array
return [[(os.path.basename(v[0][0]),) + v[0][1:], "..."] if len(v) > 1
else [(os.path.basename(v[0][0]),) + v[0][1:]] for v in value]
elif type_ == "LoRA Batch" and isinstance(value, list):
# Extract the basename of the first value of the first tuple from each sublist
return [os.path.basename(v[0][0]) for v in value if v and isinstance(v[0], tuple) and v[0][0]]
elif (type_ == "LoRA Wt" or type_ == "LoRA MStr") and isinstance(value, list):
# Extract the first value of the first tuple from each sublist
return [v[0][1] for v in value if v and isinstance(v[0], tuple)]
elif type_ == "LoRA CStr" and isinstance(value, list):
# Extract the first value of the first tuple from each sublist
return [v[0][2] for v in value if v and isinstance(v[0], tuple)]
elif type_ == "ControlNetStrength" and isinstance(value, list):
# Extract the third entry of the first tuple from each inner list
return [round(inner_list[0][2], 3) for inner_list in value]
elif type_ == "ControlNetStart%" and isinstance(value, list):
# Extract the third entry of the first tuple from each inner list
return [round(inner_list[0][3], 3) for inner_list in value]
elif type_ == "ControlNetEnd%" and isinstance(value, list):
# Extract the third entry of the first tuple from each inner list
return [round(inner_list[0][4], 3) for inner_list in value]
elif isinstance(value, tuple):
return tuple(replacement if v is None else v for v in value)
else:
return replacement if value is None else value
# Determine the replacements based on X_type and Y_type
replacement_X = scheduler if X_type == 'Sampler' else clip_skip if X_type == 'Checkpoint' else None
replacement_Y = scheduler if Y_type == 'Sampler' else clip_skip if Y_type == 'Checkpoint' else None
# Process X_value and Y_value
X_value_processed = process_xy_for_print(X_value, replacement_X, X_type)
Y_value_processed = process_xy_for_print(Y_value, replacement_Y, Y_type)
print(info("-" * 40))
print(info('XY Plot Script Inputs:'))
print(info(f"(X) {X_type}:"))
for item in X_value_processed:
print(info(f" {item}"))
print(info(f"(Y) {Y_type}:"))
for item in Y_value_processed:
print(info(f" {item}"))
print(info("-" * 40))
#_______________________________________________________________________________________________________
# Perform various initializations in this section
# If not caching models, set to 1.
if cache_models == "False":
vae_cache = ckpt_cache = lora_cache = refn_cache = 1
else:
# Retrieve cache numbers
vae_cache, ckpt_cache, lora_cache, refn_cache = get_cache_numbers("XY Plot")
# Pack cache numbers in a tuple
cache = (vae_cache, ckpt_cache, lora_cache, refn_cache)
# Add seed to every entry in the list
X_value = [v + seed for v in X_value] if "Seeds++ Batch" == X_type else X_value
Y_value = [v + seed for v in Y_value] if "Seeds++ Batch" == Y_type else Y_value
# Embedd original prompts into prompt variables
positive_prompt = (positive_prompt, positive_prompt)
negative_prompt = (negative_prompt, negative_prompt)
# Set lora_stack to None if one of types are LoRA
if "LoRA" in X_type or "LoRA" in Y_type:
lora_stack = None
# Define the manipulated and static Control Net Variables with a tuple with shape (cn_1, cn_2, cn_3).
# The information in this tuple will be used by the plotter to properly plot Control Net XY input types.
cn_1, cn_2, cn_3 = None, None, None
# If X_type has "ControlNet" or both X_type and Y_type have "ControlNet"
if "ControlNet" in X_type:
cn_1, cn_2, cn_3 = X_value[0][0][2], X_value[0][0][3], X_value[0][0][4]
# If only Y_type has "ControlNet" and not X_type
elif "ControlNet" in Y_type:
cn_1, cn_2, cn_3 = Y_value[0][0][2], Y_value[0][0][3], Y_value[0][0][4]
# Additional checks for other substrings
if "ControlNetStrength" in X_type or "ControlNetStrength" in Y_type:
cn_1 = None
if "ControlNetStart%" in X_type or "ControlNetStart%" in Y_type:
cn_2 = None
if "ControlNetEnd%" in X_type or "ControlNetEnd%" in Y_type:
cn_3 = None
# Embed the information in cnet_stack
cnet_stack = (cnet_stack, (cn_1, cn_2, cn_3))
# Optimize image generation by prioritization:
priority = [
"Checkpoint",
"Refiner",
"LoRA",
"VAE",
]
conditioners = {
"Positive Prompt S/R",
"Negative Prompt S/R",
"AScore+",
"AScore-",
"Clip Skip",
"Clip Skip (Refiner)",
"ControlNetStrength",
"ControlNetStart%",
"ControlNetEnd%"
}
# Get priority values; return a high number if the type is not in priority list
x_priority = priority.index(X_type) if X_type in priority else 999
y_priority = priority.index(Y_type) if Y_type in priority else 999
# Check if both are conditioners
are_both_conditioners = X_type in conditioners and Y_type in conditioners
# Special cases
is_special_case = (
(X_type == "Refiner On/Off" and Y_type in ["RefineStep", "Steps"]) or
(X_type == "Nothing" and Y_type != "Nothing")
)
# Determine whether to flip
flip_xy = (y_priority < x_priority and not are_both_conditioners) or is_special_case
# Perform the flip if necessary
if flip_xy:
X_type, Y_type = Y_type, X_type
X_value, Y_value = Y_value, X_value
#_______________________________________________________________________________________________________
# The below code will clean from the cache any ckpt/vae/lora models it will not be reusing.
# Note: Special LoRA types will not trigger cache: "LoRA Batch", "LoRA Wt", "LoRA MStr", "LoRA CStr"
# Map the type names to the dictionaries
dict_map = {"VAE": [], "Checkpoint": [], "LoRA": [], "Refiner": []}
# Create a list of tuples with types and values
type_value_pairs = [(X_type, X_value.copy()), (Y_type, Y_value.copy())]
# Iterate over type-value pairs
for t, v in type_value_pairs:
if t in dict_map:
# Flatten the list of lists of tuples if the type is "LoRA"
if t == "LoRA":
dict_map[t] = [item for sublist in v for item in sublist]
else:
dict_map[t] = v
vae_dict = dict_map.get("VAE", [])
# Construct ckpt_dict and also update vae_dict based on the third entry of the tuples in dict_map["Checkpoint"]
if dict_map.get("Checkpoint", []):
ckpt_dict = [t[0] for t in dict_map["Checkpoint"]]
for t in dict_map["Checkpoint"]:
if t[2] is not None and t[2] != "Baked VAE":
vae_dict.append(t[2])
else:
ckpt_dict = []
lora_dict = [[t,] for t in dict_map.get("LoRA", [])] if dict_map.get("LoRA", []) else []
# Construct refn_dict
if dict_map.get("Refiner", []):
refn_dict = [t[0] for t in dict_map["Refiner"]]
else:
refn_dict = []
# If both ckpt_dict and lora_dict are not empty, manipulate lora_dict as described
if ckpt_dict and lora_dict:
lora_dict = [(lora_stack, ckpt) for ckpt in ckpt_dict for lora_stack in lora_dict]
# If lora_dict is not empty and ckpt_dict is empty, insert ckpt_name into each tuple in lora_dict
elif lora_dict:
lora_dict = [(lora_stack, ckpt_name) for lora_stack in lora_dict]
# Avoid caching models accross both X and Y
if X_type == "Checkpoint":
lora_dict = []
refn_dict = []
elif X_type == "Refiner":
ckpt_dict = []
lora_dict = []
elif X_type == "LoRA":
ckpt_dict = []
refn_dict = []
### Print dict_arrays for debugging
###print(f"vae_dict={vae_dict}\nckpt_dict={ckpt_dict}\nlora_dict={lora_dict}\nrefn_dict={refn_dict}")
# Clean values that won't be reused
clear_cache_by_exception(xyplot_id, vae_dict=vae_dict, ckpt_dict=ckpt_dict, lora_dict=lora_dict, refn_dict=refn_dict)
### Print loaded_objects for debugging
###print_loaded_objects_entries()
#_______________________________________________________________________________________________________
# Function that changes appropiate variables for next processed generations (also generates XY_labels)
def define_variable(var_type, var, add_noise, seed, steps, start_at_step, end_at_step,
return_with_leftover_noise, cfg, sampler_name, scheduler, denoise, vae_name, ckpt_name,
clip_skip, refiner_name, refiner_clip_skip, positive_prompt, negative_prompt, ascore,
lora_stack, cnet_stack, var_label, num_label):
# Define default max label size limit
max_label_len = 42
# If var_type is "AddNoise", update 'add_noise' with 'var', and generate text label
if var_type == "AddNoise":
add_noise = var
text = f"AddNoise: {add_noise}"
# If var_type is "Seeds++ Batch", generate text label
elif var_type == "Seeds++ Batch":
seed = var
text = f"Seed: {seed}"
# If var_type is "Steps", update 'steps' with 'var' and generate text label
elif var_type == "Steps":
steps = var
text = f"Steps: {steps}"
# If var_type is "StartStep", update 'start_at_step' with 'var' and generate text label
elif var_type == "StartStep":
start_at_step = var
text = f"StartStep: {start_at_step}"
# If var_type is "EndStep", update 'end_at_step' with 'var' and generate text label
elif var_type == "EndStep":
end_at_step = var
text = f"EndStep: {end_at_step}"
# If var_type is "RefineStep", update 'end_at_step' with 'var' and generate text label
elif var_type == "RefineStep":
end_at_step = var
text = f"RefineStep: {end_at_step}"
# If var_type is "ReturnNoise", update 'return_with_leftover_noise' with 'var', and generate text label
elif var_type == "ReturnNoise":
return_with_leftover_noise = var
text = f"ReturnNoise: {return_with_leftover_noise}"
# If var_type is "CFG Scale", update cfg with var and generate text label
elif var_type == "CFG Scale":
cfg = var
text = f"CFG: {round(cfg,2)}"
# If var_type is "Sampler", update sampler_name and scheduler with var, and generate text label
elif var_type == "Sampler":
sampler_name = var[0]
if var[1] == "":
text = f"{sampler_name}"
else:
if var[1] != None:
scheduler = (var[1], scheduler[1])
else:
scheduler = (scheduler[1], scheduler[1])
text = f"{sampler_name} ({scheduler[0]})"
text = text.replace("ancestral", "a").replace("uniform", "u").replace("exponential","exp")
# If var_type is "Scheduler", update scheduler and generate labels
elif var_type == "Scheduler":
if len(var) == 2:
scheduler = (var[0], scheduler[1])
text = f"{sampler_name} ({scheduler[0]})"
else:
scheduler = (var, scheduler[1])
text = f"{scheduler[0]}"
text = text.replace("ancestral", "a").replace("uniform", "u").replace("exponential","exp")
# If var_type is "Denoise", update denoise and generate labels
elif var_type == "Denoise":
denoise = var
text = f"Denoise: {round(denoise, 2)}"
# If var_type is "VAE", update vae_name and generate labels
elif var_type == "VAE":
vae_name = var
vae_filename = os.path.splitext(os.path.basename(vae_name))[0]
text = f"VAE: {vae_filename}"
# If var_type is "Positive Prompt S/R", update positive_prompt and generate labels
elif var_type == "Positive Prompt S/R":
search_txt, replace_txt = var
if replace_txt != None:
# check if we are in the Y loop after the X loop
if positive_prompt[2] is not None:
positive_prompt = (positive_prompt[2].replace(search_txt, replace_txt, 1), positive_prompt[1], positive_prompt[2])
else:
positive_prompt = (positive_prompt[1].replace(search_txt, replace_txt, 1), positive_prompt[1], positive_prompt[1].replace(search_txt, replace_txt, 1))
else:
if positive_prompt[2] is not None:
positive_prompt = (positive_prompt[2], positive_prompt[1], positive_prompt[2])
else:
positive_prompt = (positive_prompt[1], positive_prompt[1], positive_prompt[1])
replace_txt = search_txt
text = f"{replace_txt}"
# If var_type is "Negative Prompt S/R", update negative_prompt and generate labels
elif var_type == "Negative Prompt S/R":
search_txt, replace_txt = var
if replace_txt != None:
# check if we are in the Y loop after the X loop
if negative_prompt[2] is not None:
negative_prompt = (negative_prompt[2].replace(search_txt, replace_txt, 1), negative_prompt[1], negative_prompt[2])
else:
negative_prompt = (negative_prompt[1].replace(search_txt, replace_txt, 1), negative_prompt[1], negative_prompt[1].replace(search_txt, replace_txt, 1))
else:
if negative_prompt[2] is not None:
negative_prompt = (negative_prompt[2], negative_prompt[1], negative_prompt[2])
else:
negative_prompt = (negative_prompt[1], negative_prompt[1], negative_prompt[1])
replace_txt = search_txt
text = f"(-) {replace_txt}"
# If var_type is "AScore+", update positive ascore and generate labels
elif var_type == "AScore+":
ascore = (var,ascore[1])
text = f"+AScore: {ascore[0]}"
# If var_type is "AScore-", update negative ascore and generate labels
elif var_type == "AScore-":
ascore = (ascore[0],var)
text = f"-AScore: {ascore[1]}"
# If var_type is "Checkpoint", update model and clip (if needed) and generate labels
elif var_type == "Checkpoint":
ckpt_name = var[0]
if var[1] == None:
clip_skip = (clip_skip[1],clip_skip[1])
else:
clip_skip = (var[1],clip_skip[1])
if var[2] != None:
vae_name = var[2]
ckpt_filename = os.path.splitext(os.path.basename(ckpt_name))[0]
text = f"{ckpt_filename}"
# If var_type is "Refiner", update model and clip (if needed) and generate labels
elif var_type == "Refiner":
refiner_name = var[0]
if var[1] == None:
refiner_clip_skip = (refiner_clip_skip[1],refiner_clip_skip[1])
else:
refiner_clip_skip = (var[1],refiner_clip_skip[1])
ckpt_filename = os.path.splitext(os.path.basename(refiner_name))[0]
text = f"{ckpt_filename}"
# If var_type is "Refiner On/Off", set end_at_step = max steps and generate labels
elif var_type == "Refiner On/Off":
end_at_step = int(var * steps)
text = f"Refiner: {'On' if var < 1 else 'Off'}"
elif var_type == "Clip Skip":
clip_skip = (var, clip_skip[1])
text = f"ClipSkip ({clip_skip[0]})"
elif var_type == "Clip Skip (Refiner)":
refiner_clip_skip = (var, refiner_clip_skip[1])
text = f"RefClipSkip ({refiner_clip_skip[0]})"
elif "LoRA" in var_type:
if not lora_stack:
lora_stack = var.copy()
else:
# Updating the first tuple of lora_stack
lora_stack[0] = tuple(v if v is not None else lora_stack[0][i] for i, v in enumerate(var[0]))
max_label_len = 50 + (12 * (len(lora_stack) - 1))
lora_name, lora_model_wt, lora_clip_wt = lora_stack[0]
lora_filename = os.path.splitext(os.path.basename(lora_name))[0]
if var_type == "LoRA":
if len(lora_stack) == 1:
lora_model_wt = format(float(lora_model_wt), ".2f").rstrip('0').rstrip('.')
lora_clip_wt = format(float(lora_clip_wt), ".2f").rstrip('0').rstrip('.')
lora_filename = lora_filename[:max_label_len - len(f"LoRA: ({lora_model_wt})")]
if lora_model_wt == lora_clip_wt:
text = f"LoRA: {lora_filename}({lora_model_wt})"
else:
text = f"LoRA: {lora_filename}({lora_model_wt},{lora_clip_wt})"
elif len(lora_stack) > 1:
lora_filenames = [os.path.splitext(os.path.basename(lora_name))[0] for lora_name, _, _ in
lora_stack]
lora_details = [(format(float(lora_model_wt), ".2f").rstrip('0').rstrip('.'),
format(float(lora_clip_wt), ".2f").rstrip('0').rstrip('.')) for
_, lora_model_wt, lora_clip_wt in lora_stack]
non_name_length = sum(
len(f"({lora_details[i][0]},{lora_details[i][1]})") + 2 for i in range(len(lora_stack)))
available_space = max_label_len - non_name_length
max_name_length = available_space // len(lora_stack)
lora_filenames = [filename[:max_name_length] for filename in lora_filenames]
text_elements = [
f"{lora_filename}({lora_details[i][0]})" if lora_details[i][0] == lora_details[i][1]
else f"{lora_filename}({lora_details[i][0]},{lora_details[i][1]})" for i, lora_filename in
enumerate(lora_filenames)]
text = " ".join(text_elements)
elif var_type == "LoRA Batch":
text = f"LoRA: {lora_filename}"
elif var_type == "LoRA Wt":
lora_model_wt = format(float(lora_model_wt), ".2f").rstrip('0').rstrip('.')
text = f"LoRA Wt: {lora_model_wt}"
elif var_type == "LoRA MStr":
lora_model_wt = format(float(lora_model_wt), ".2f").rstrip('0').rstrip('.')
text = f"LoRA Mstr: {lora_model_wt}"
elif var_type == "LoRA CStr":
lora_clip_wt = format(float(lora_clip_wt), ".2f").rstrip('0').rstrip('.')
text = f"LoRA Cstr: {lora_clip_wt}"
elif var_type in ["ControlNetStrength", "ControlNetStart%", "ControlNetEnd%"]:
if "Strength" in var_type:
entry_index = 2
elif "Start%" in var_type:
entry_index = 3
elif "End%" in var_type:
entry_index = 4
# If the first entry of cnet_stack is None, set it to var
if cnet_stack[0] is None:
cnet_stack = (var, cnet_stack[1])
else:
# Extract the desired entry from var's first tuple
entry_from_var = var[0][entry_index]
# Extract the first tuple from cnet_stack[0][0] and make it mutable
first_cn_entry = list(cnet_stack[0][0])
# Replace the appropriate entry
first_cn_entry[entry_index] = entry_from_var
# Further update first_cn_entry based on cnet_stack[1]
for i, value in enumerate(cnet_stack[1][-3:]): # Considering last 3 entries
if value is not None:
first_cn_entry[i + 2] = value # "+2" to offset for the first 2 entries of the tuple
# Convert back to tuple for the updated values
updated_first_entry = tuple(first_cn_entry)
# Construct the updated cnet_stack[0] using the updated_first_entry and the rest of the values from cnet_stack[0]
updated_cnet_stack_0 = [updated_first_entry] + list(cnet_stack[0][1:])
# Update cnet_stack
cnet_stack = (updated_cnet_stack_0, cnet_stack[1])
# Print the desired value
text = f'{var_type}: {round(cnet_stack[0][0][entry_index], 3)}'
elif var_type == "XY_Capsule":
text = var.getLabel()
else: # No matching type found
text=""
def truncate_texts(texts, num_label, max_label_len):
truncate_length = max(min(max(len(text) for text in texts), max_label_len), 24)
return [text if len(text) <= truncate_length else text[:truncate_length] + "..." for text in
texts]
# Add the generated text to var_label if it's not full
if len(var_label) < num_label:
var_label.append(text)
# If var_type VAE , truncate entries in the var_label list when it's full
if len(var_label) == num_label and (var_type == "VAE" or var_type == "Checkpoint"
or var_type == "Refiner" or "LoRA" in var_type):
var_label = truncate_texts(var_label, num_label, max_label_len)
# Return the modified variables
return add_noise, seed, steps, start_at_step, end_at_step, return_with_leftover_noise, cfg,\
sampler_name, scheduler, denoise, vae_name, ckpt_name, clip_skip, \
refiner_name, refiner_clip_skip, positive_prompt, negative_prompt, ascore,\
lora_stack, cnet_stack, var_label
#_______________________________________________________________________________________________________
# The function below is used to optimally load Checkpoint/LoRA/VAE models between generations.
def define_model(model, clip, clip_skip, refiner_model, refiner_clip, refiner_clip_skip,
ckpt_name, refiner_name, positive, negative, refiner_positive, refiner_negative,
positive_prompt, negative_prompt, ascore, vae, vae_name, lora_stack, cnet_stack, index,
types, xyplot_id, cache, sampler_type, empty_latent_width, empty_latent_height):
# Variable to track wether to encode prompt or not
encode = False
encode_refiner = False
# Unpack types tuple
X_type, Y_type = types
# Note: Index is held at 0 when Y_type == "Nothing"
# Load Checkpoint if required. If Y_type is LoRA, required models will be loaded by load_lora func.
if (X_type == "Checkpoint" and index == 0 and Y_type != "LoRA"):
if lora_stack is None:
model, clip, _ = load_checkpoint(ckpt_name, xyplot_id, cache=cache[1])
else: # Load Efficient Loader LoRA
model, clip = load_lora(lora_stack, ckpt_name, xyplot_id,
cache=None, ckpt_cache=cache[1])
encode = True
# Load LoRA if required
elif (X_type == "LoRA"):
# Don't cache Checkpoints
model, clip = load_lora(lora_stack, ckpt_name, xyplot_id, cache=cache[2])
encode = True
elif Y_type == "LoRA": # X_type must be Checkpoint, so cache those as defined
model, clip = load_lora(lora_stack, ckpt_name, xyplot_id,
cache=None, ckpt_cache=cache[1])
encode = True
elif X_type == "LoRA Batch" or X_type == "LoRA Wt" or X_type == "LoRA MStr" or X_type == "LoRA CStr":
# Don't cache Checkpoints or LoRAs
model, clip = load_lora(lora_stack, ckpt_name, xyplot_id, cache=0)
encode = True
if (X_type == "Refiner" and index == 0) or Y_type == "Refiner":
refiner_model, refiner_clip, _ = \
load_checkpoint(refiner_name, xyplot_id, output_vae=False, cache=cache[3], ckpt_type="refn")
encode_refiner = True
# Encode base prompt if required
encode_types = ["Positive Prompt S/R", "Negative Prompt S/R", "Clip Skip", "ControlNetStrength",
"ControlNetStart%", "ControlNetEnd%", "XY_Capsule"]
if (X_type in encode_types and index == 0) or Y_type in encode_types:
encode = True
# Encode refiner prompt if required
encode_refiner_types = ["Positive Prompt S/R", "Negative Prompt S/R", "AScore+", "AScore-",
"Clip Skip (Refiner)", "XY_Capsule"]
if (X_type in encode_refiner_types and index == 0) or Y_type in encode_refiner_types:
encode_refiner = True
# Encode base prompt
if encode == True:
positive, negative, clip = \
encode_prompts(positive_prompt, negative_prompt, token_normalization, weight_interpretation,
clip, clip_skip, refiner_clip, refiner_clip_skip, ascore, sampler_type == "sdxl",
empty_latent_width, empty_latent_height, return_type="base")
# Apply ControlNet Stack if given
if cnet_stack:
controlnet_conditioning = TSC_Apply_ControlNet_Stack().apply_cnet_stack(positive, negative, cnet_stack)
positive, negative = controlnet_conditioning[0], controlnet_conditioning[1]
if encode_refiner == True:
refiner_positive, refiner_negative, refiner_clip = \
encode_prompts(positive_prompt, negative_prompt, token_normalization, weight_interpretation,
clip, clip_skip, refiner_clip, refiner_clip_skip, ascore, sampler_type == "sdxl",
empty_latent_width, empty_latent_height, return_type="refiner")
# Load VAE if required
if (X_type == "VAE" and index == 0) or Y_type == "VAE":
#vae = load_vae(vae_name, xyplot_id, cache=cache[0])
vae = get_bvae_by_ckpt_name(ckpt_name) if vae_name == "Baked VAE" \
else load_vae(vae_name, xyplot_id, cache=cache[0])
elif X_type == "Checkpoint" and index == 0 and vae_name:
vae = get_bvae_by_ckpt_name(ckpt_name) if vae_name == "Baked VAE" \
else load_vae(vae_name, xyplot_id, cache=cache[0])
return model, positive, negative, refiner_model, refiner_positive, refiner_negative, vae
# ______________________________________________________________________________________________________
# The below function is used to generate the results based on all the processed variables
def process_values(model, refiner_model, add_noise, seed, steps, start_at_step, end_at_step,
return_with_leftover_noise, cfg, sampler_name, scheduler, positive, negative,
refiner_positive, refiner_negative, latent_image, denoise, vae, vae_decode,
sampler_type, latent_list=[], image_tensor_list=[], image_pil_list=[], xy_capsule=None):
capsule_result = None
if xy_capsule is not None:
capsule_result = xy_capsule.get_result(model, clip, vae)
if capsule_result is not None:
image, latent = capsule_result
latent_list.append(latent)
if capsule_result is None:
samples, images, _, _ = process_latent_image(model, seed, steps, cfg, sampler_name, scheduler, positive, negative,
latent_image, denoise, sampler_type, add_noise, start_at_step,
end_at_step, return_with_leftover_noise, refiner_model,
refiner_positive, refiner_negative, vae, vae_decode, preview_method)
# Add the latent tensor to the tensors list
latent_list.append(samples)
# Decode the latent tensor if required
image = images if images is not None else vae_decode_latent(vae, samples, vae_decode)
if xy_capsule is not None:
xy_capsule.set_result(image, samples)
# Add the resulting image tensor to image_tensor_list
image_tensor_list.append(image)
# Convert the image from tensor to PIL Image and add it to the image_pil_list
image_pil_list.append(tensor2pil(image))
# Return the touched variables
return latent_list, image_tensor_list, image_pil_list
# ______________________________________________________________________________________________________
# The below section is the heart of the XY Plot image generation
# Initiate Plot label text variables X/Y_label
X_label = []
Y_label = []
# Store the KSamplers original scheduler inside the same scheduler variable
scheduler = (scheduler, scheduler)
# Store the Eff Loaders original clip_skips inside the same clip_skip variables
clip_skip = (clip_skip, clip_skip)
refiner_clip_skip = (refiner_clip_skip, refiner_clip_skip)
# Store types in a Tuple for easy function passing
types = (X_type, Y_type)
# Clone original model parameters
def clone_or_none(*originals):
cloned_items = []
for original in originals:
try:
cloned_items.append(original.clone())
except (AttributeError, TypeError):
# If not clonable, just append the original item
cloned_items.append(original)
return cloned_items
original_model, original_clip, original_positive, original_negative,\
original_refiner_model, original_refiner_clip, original_refiner_positive, original_refiner_negative =\
clone_or_none(model, clip, positive, negative, refiner_model, refiner_clip, refiner_positive, refiner_negative)
# Fill Plot Rows (X)
for X_index, X in enumerate(X_value):
# add a none value in the positive prompt memory.
# the tuple is composed of (actual prompt, original prompte before S/R, prompt after X S/R)
positive_prompt = (positive_prompt[0], positive_prompt[1], None)
negative_prompt = (negative_prompt[0], negative_prompt[1], None)
# Define X parameters and generate labels
add_noise, seed, steps, start_at_step, end_at_step, return_with_leftover_noise, cfg,\
sampler_name, scheduler, denoise, vae_name, ckpt_name, clip_skip,\
refiner_name, refiner_clip_skip, positive_prompt, negative_prompt, ascore,\
lora_stack, cnet_stack, X_label = \
define_variable(X_type, X, add_noise, seed, steps, start_at_step, end_at_step,
return_with_leftover_noise, cfg, sampler_name, scheduler, denoise, vae_name,
ckpt_name, clip_skip, refiner_name, refiner_clip_skip, positive_prompt,
negative_prompt, ascore, lora_stack, cnet_stack, X_label, len(X_value))
if X_type != "Nothing" and Y_type == "Nothing":
if X_type == "XY_Capsule":
model, clip, refiner_model, refiner_clip = \
clone_or_none(original_model, original_clip, original_refiner_model, original_refiner_clip)
model, clip, vae = X.pre_define_model(model, clip, vae)
# Models & Conditionings
model, positive, negative, refiner_model, refiner_positive, refiner_negative, vae = \
define_model(model, clip, clip_skip[0], refiner_model, refiner_clip, refiner_clip_skip[0],
ckpt_name, refiner_name, positive, negative, refiner_positive, refiner_negative,
positive_prompt[0], negative_prompt[0], ascore, vae, vae_name, lora_stack, cnet_stack[0],
0, types, xyplot_id, cache, sampler_type, empty_latent_width, empty_latent_height)
xy_capsule = None
if X_type == "XY_Capsule":
xy_capsule = X
# Generate Results
latent_list, image_tensor_list, image_pil_list = \
process_values(model, refiner_model, add_noise, seed, steps, start_at_step, end_at_step,
return_with_leftover_noise, cfg, sampler_name, scheduler[0], positive, negative,
refiner_positive, refiner_negative, latent_image, denoise, vae, vae_decode, sampler_type, xy_capsule=xy_capsule)
elif X_type != "Nothing" and Y_type != "Nothing":
for Y_index, Y in enumerate(Y_value):
if Y_type == "XY_Capsule" or X_type == "XY_Capsule":
model, clip, refiner_model, refiner_clip = \
clone_or_none(original_model, original_clip, original_refiner_model, original_refiner_clip)
if Y_type == "XY_Capsule" and X_type == "XY_Capsule":
Y.set_x_capsule(X)
# Define Y parameters and generate labels
add_noise, seed, steps, start_at_step, end_at_step, return_with_leftover_noise, cfg,\
sampler_name, scheduler, denoise, vae_name, ckpt_name, clip_skip,\
refiner_name, refiner_clip_skip, positive_prompt, negative_prompt, ascore,\
lora_stack, cnet_stack, Y_label = \
define_variable(Y_type, Y, add_noise, seed, steps, start_at_step, end_at_step,
return_with_leftover_noise, cfg, sampler_name, scheduler, denoise, vae_name,
ckpt_name, clip_skip, refiner_name, refiner_clip_skip, positive_prompt,
negative_prompt, ascore, lora_stack, cnet_stack, Y_label, len(Y_value))
if Y_type == "XY_Capsule":
model, clip, vae = Y.pre_define_model(model, clip, vae)
elif X_type == "XY_Capsule":
model, clip, vae = X.pre_define_model(model, clip, vae)
# Models & Conditionings
model, positive, negative, refiner_model, refiner_positive, refiner_negative, vae = \
define_model(model, clip, clip_skip[0], refiner_model, refiner_clip, refiner_clip_skip[0],
ckpt_name, refiner_name, positive, negative, refiner_positive, refiner_negative,
positive_prompt[0], negative_prompt[0], ascore, vae, vae_name, lora_stack, cnet_stack[0],
Y_index, types, xyplot_id, cache, sampler_type, empty_latent_width,
empty_latent_height)
# Generate Results
xy_capsule = None
if Y_type == "XY_Capsule":
xy_capsule = Y
latent_list, image_tensor_list, image_pil_list = \
process_values(model, refiner_model, add_noise, seed, steps, start_at_step, end_at_step,
return_with_leftover_noise, cfg, sampler_name, scheduler[0],
positive, negative, refiner_positive, refiner_negative, latent_image,
denoise, vae, vae_decode, sampler_type, xy_capsule=xy_capsule)
# Clean up cache
if cache_models == "False":
clear_cache_by_exception(xyplot_id, vae_dict=[], ckpt_dict=[], lora_dict=[], refn_dict=[])
else:
# Avoid caching models accross both X and Y
if X_type == "Checkpoint":
clear_cache_by_exception(xyplot_id, lora_dict=[], refn_dict=[])
elif X_type == "Refiner":
clear_cache_by_exception(xyplot_id, ckpt_dict=[], lora_dict=[])
elif X_type == "LoRA":
clear_cache_by_exception(xyplot_id, ckpt_dict=[], refn_dict=[])
# __________________________________________________________________________________________________________
# Function for printing all plot variables (WARNING: This function is an absolute mess)
def print_plot_variables(X_type, Y_type, X_value, Y_value, add_noise, seed, steps, start_at_step, end_at_step,
return_with_leftover_noise, cfg, sampler_name, scheduler, denoise, vae_name, ckpt_name,
clip_skip, refiner_name, refiner_clip_skip, ascore, lora_stack, cnet_stack, sampler_type,
num_rows, num_cols, i_height, i_width):
print("-" * 40) # Print an empty line followed by a separator line
print(f"{xyplot_message('XY Plot Results:')}")
def get_vae_name(X_type, Y_type, X_value, Y_value, vae_name):
if X_type == "VAE":
vae_name = "\n ".join(map(lambda x: os.path.splitext(os.path.basename(str(x)))[0], X_value))
elif Y_type == "VAE":
vae_name = "\n ".join(map(lambda y: os.path.splitext(os.path.basename(str(y)))[0], Y_value))
elif vae_name:
vae_name = os.path.splitext(os.path.basename(str(vae_name)))[0]
else:
vae_name = ""
return vae_name
def get_clip_skip(X_type, Y_type, X_value, Y_value, cskip, mode):
clip_type = "Clip Skip" if mode == "ckpt" else "Clip Skip (Refiner)"
if X_type == clip_type:
cskip = ", ".join(map(str, X_value))
elif Y_type == clip_type:
cskip = ", ".join(map(str, Y_value))
elif cskip[1] != None:
cskip = cskip[1]
else:
cskip = ""
return cskip
def get_checkpoint_name(X_type, Y_type, X_value, Y_value, ckpt_name, clip_skip, mode, vae_name=None):
# If ckpt_name is None, return it as is
if ckpt_name is not None:
ckpt_name = os.path.basename(ckpt_name)
# Define types based on mode
primary_type = "Checkpoint" if mode == "ckpt" else "Refiner"
clip_type = "Clip Skip" if mode == "ckpt" else "Clip Skip (Refiner)"
# Determine ckpt and othr based on primary type
if X_type == primary_type:
ckpt_type, ckpt_value = X_type, X_value.copy()
othr_type, othr_value = Y_type, Y_value.copy()
elif Y_type == primary_type:
ckpt_type, ckpt_value = Y_type, Y_value.copy()
othr_type, othr_value = X_type, X_value.copy()
else:
# Process as per original function if mode is "ckpt"
clip_skip = get_clip_skip(X_type, Y_type, X_value, Y_value, clip_skip, mode)
if mode == "ckpt":
if vae_name:
vae_name = get_vae_name(X_type, Y_type, X_value, Y_value, vae_name)
return ckpt_name, clip_skip, vae_name
else:
# For refn mode
return ckpt_name, clip_skip
# Process clip skip based on mode
if othr_type == clip_type:
clip_skip = ", ".join(map(str, othr_value))
elif ckpt_value[0][1] != None:
clip_skip = None
# Process vae_name based on mode
if mode == "ckpt":
if othr_type == "VAE":
vae_name = get_vae_name(X_type, Y_type, X_value, Y_value, vae_name)
elif ckpt_value[0][2] != None:
vae_name = None
def format_name(v, _type):
base = os.path.basename(v[0])
if _type == clip_type and v[1] is not None:
return base
elif _type == "VAE" and v[1] is not None and v[2] is not None:
return f"{base}({v[1]})"
elif v[1] is not None and v[2] is not None:
return f"{base}({v[1]}) + vae:{v[2]}"
elif v[1] is not None:
return f"{base}({v[1]})"
else:
return base
ckpt_name = "\n ".join([format_name(v, othr_type) for v in ckpt_value])
if mode == "ckpt":
return ckpt_name, clip_skip, vae_name
else:
return ckpt_name, clip_skip
def get_lora_name(X_type, Y_type, X_value, Y_value, lora_stack=None):
lora_name = lora_wt = lora_model_str = lora_clip_str = None
# Check for all possible LoRA types
lora_types = ["LoRA", "LoRA Batch", "LoRA Wt", "LoRA MStr", "LoRA CStr"]
if X_type not in lora_types and Y_type not in lora_types:
if lora_stack:
names_list = []
for name, model_wt, clip_wt in lora_stack:
base_name = os.path.splitext(os.path.basename(name))[0]
formatted_str = f"{base_name}({round(model_wt, 3)},{round(clip_wt, 3)})"
names_list.append(formatted_str)
lora_name = f"[{', '.join(names_list)}]"
else:
if X_type in lora_types:
value = get_lora_sublist_name(X_type, X_value)
if X_type == "LoRA":
lora_name = value
lora_model_str = None
lora_clip_str = None
if X_type == "LoRA Batch":
lora_name = value
lora_model_str = X_value[0][0][1] if lora_model_str is None else lora_model_str
lora_clip_str = X_value[0][0][2] if lora_clip_str is None else lora_clip_str
elif X_type == "LoRA MStr":
lora_name = os.path.basename(X_value[0][0][0]) if lora_name is None else lora_name
lora_model_str = value
lora_clip_str = X_value[0][0][2] if lora_clip_str is None else lora_clip_str
elif X_type == "LoRA CStr":
lora_name = os.path.basename(X_value[0][0][0]) if lora_name is None else lora_name
lora_model_str = X_value[0][0][1] if lora_model_str is None else lora_model_str
lora_clip_str = value
elif X_type == "LoRA Wt":
lora_name = os.path.basename(X_value[0][0][0]) if lora_name is None else lora_name
lora_wt = value
if Y_type in lora_types:
value = get_lora_sublist_name(Y_type, Y_value)
if Y_type == "LoRA":
lora_name = value
lora_model_str = None
lora_clip_str = None
if Y_type == "LoRA Batch":
lora_name = value
lora_model_str = Y_value[0][0][1] if lora_model_str is None else lora_model_str
lora_clip_str = Y_value[0][0][2] if lora_clip_str is None else lora_clip_str
elif Y_type == "LoRA MStr":
lora_name = os.path.basename(Y_value[0][0][0]) if lora_name is None else lora_name
lora_model_str = value
lora_clip_str = Y_value[0][0][2] if lora_clip_str is None else lora_clip_str
elif Y_type == "LoRA CStr":
lora_name = os.path.basename(Y_value[0][0][0]) if lora_name is None else lora_name
lora_model_str = Y_value[0][0][1] if lora_model_str is None else lora_model_str
lora_clip_str = value
elif Y_type == "LoRA Wt":
lora_name = os.path.basename(Y_value[0][0][0]) if lora_name is None else lora_name
lora_wt = value
return lora_name, lora_wt, lora_model_str, lora_clip_str
def get_lora_sublist_name(lora_type, lora_value):
if lora_type == "LoRA" or lora_type == "LoRA Batch":
formatted_sublists = []
for sublist in lora_value:
formatted_entries = []
for x in sublist:
base_name = os.path.splitext(os.path.basename(str(x[0])))[0]
formatted_str = f"{base_name}({round(x[1], 3)},{round(x[2], 3)})" if lora_type == "LoRA" else f"{base_name}"
formatted_entries.append(formatted_str)
formatted_sublists.append(f"{', '.join(formatted_entries)}")
return "\n ".join(formatted_sublists)
elif lora_type == "LoRA MStr":
return ", ".join([str(round(x[0][1], 3)) for x in lora_value])
elif lora_type == "LoRA CStr":
return ", ".join([str(round(x[0][2], 3)) for x in lora_value])
elif lora_type == "LoRA Wt":
return ", ".join([str(round(x[0][1], 3)) for x in lora_value]) # assuming LoRA Wt uses the second value
else:
return ""
# VAE, Checkpoint, Clip Skip, LoRA
ckpt_name, clip_skip, vae_name = get_checkpoint_name(X_type, Y_type, X_value, Y_value, ckpt_name, clip_skip, "ckpt", vae_name)
lora_name, lora_wt, lora_model_str, lora_clip_str = get_lora_name(X_type, Y_type, X_value, Y_value, lora_stack)
refiner_name, refiner_clip_skip = get_checkpoint_name(X_type, Y_type, X_value, Y_value, refiner_name, refiner_clip_skip, "refn")
# AddNoise
add_noise = ", ".join(map(str, X_value)) if X_type == "AddNoise" else ", ".join(
map(str, Y_value)) if Y_type == "AddNoise" else add_noise
# Seeds++ Batch
seed = "\n ".join(map(str, X_value)) if X_type == "Seeds++ Batch" else "\n ".join(
map(str, Y_value)) if Y_type == "Seeds++ Batch" else seed
# Steps
steps = ", ".join(map(str, X_value)) if X_type == "Steps" else ", ".join(
map(str, Y_value)) if Y_type == "Steps" else steps
# StartStep
start_at_step = ", ".join(map(str, X_value)) if X_type == "StartStep" else ", ".join(
map(str, Y_value)) if Y_type == "StartStep" else start_at_step
# EndStep/RefineStep
end_at_step = ", ".join(map(str, X_value)) if X_type in ["EndStep", "RefineStep"] else ", ".join(
map(str, Y_value)) if Y_type in ["EndStep", "RefineStep"] else end_at_step
# ReturnNoise
return_with_leftover_noise = ", ".join(map(str, X_value)) if X_type == "ReturnNoise" else ", ".join(
map(str, Y_value)) if Y_type == "ReturnNoise" else return_with_leftover_noise
# CFG
cfg = ", ".join(map(str, X_value)) if X_type == "CFG Scale" else ", ".join(
map(str, Y_value)) if Y_type == "CFG Scale" else round(cfg,3)
# Sampler/Scheduler
if X_type == "Sampler":
if Y_type == "Scheduler":
sampler_name = ", ".join([f"{x[0]}" for x in X_value])
scheduler = ", ".join([f"{y}" for y in Y_value])
else:
sampler_name = ", ".join([f"{x[0]}({x[1] if x[1] != '' and x[1] is not None else scheduler[1]})" for x in X_value])
scheduler = "_"
elif Y_type == "Sampler":
if X_type == "Scheduler":
sampler_name = ", ".join([f"{y[0]}" for y in Y_value])
scheduler = ", ".join([f"{x}" for x in X_value])
else:
sampler_name = ", ".join([f"{y[0]}({y[1] if y[1] != '' and y[1] is not None else scheduler[1]})" for y in Y_value])
scheduler = "_"
else:
scheduler = ", ".join([str(x[0]) if isinstance(x, tuple) else str(x) for x in X_value]) if X_type == "Scheduler" else \
", ".join([str(y[0]) if isinstance(y, tuple) else str(y) for y in Y_value]) if Y_type == "Scheduler" else scheduler[0]
# Denoise
denoise = ", ".join(map(str, X_value)) if X_type == "Denoise" else ", ".join(
map(str, Y_value)) if Y_type == "Denoise" else round(denoise,3)
# Check if ascore is None
if ascore is None:
pos_ascore = neg_ascore = None
else:
# Ascore+
pos_ascore = (", ".join(map(str, X_value)) if X_type == "Ascore+"
else ", ".join(map(str, Y_value)) if Y_type == "Ascore+" else round(ascore[0],3))
# Ascore-
neg_ascore = (", ".join(map(str, X_value)) if X_type == "Ascore-"
else ", ".join(map(str, Y_value)) if Y_type == "Ascore-" else round(ascore[1],3))
#..........................................PRINTOUTS....................................................
print(f"(X) {X_type}")
print(f"(Y) {Y_type}")
print(f"img_count: {len(X_value)*len(Y_value)}")
print(f"img_dims: {i_height} x {i_width}")
print(f"plot_dim: {num_cols} x {num_rows}")
print(f"ckpt: {ckpt_name if ckpt_name is not None else ''}")
if clip_skip:
print(f"clip_skip: {clip_skip}")
if sampler_type == "sdxl":
if refiner_clip_skip == "_":
print(f"refiner(clipskip): {refiner_name if refiner_name is not None else ''}")
else:
print(f"refiner: {refiner_name if refiner_name is not None else ''}")
print(f"refiner_clip_skip: {refiner_clip_skip if refiner_clip_skip is not None else ''}")
print(f"+ascore: {pos_ascore if pos_ascore is not None else ''}")
print(f"-ascore: {neg_ascore if neg_ascore is not None else ''}")
if lora_name:
print(f"lora: {lora_name}")
if lora_wt:
print(f"lora_wt: {lora_wt}")
if lora_model_str:
print(f"lora_mstr: {lora_model_str}")
if lora_clip_str:
print(f"lora_cstr: {lora_clip_str}")
if vae_name:
print(f"vae: {vae_name}")
if sampler_type == "advanced":
print(f"add_noise: {add_noise}")
print(f"seed: {seed}")
print(f"steps: {steps}")
if sampler_type == "advanced":
print(f"start_at_step: {start_at_step}")
print(f"end_at_step: {end_at_step}")
print(f"return_noise: {return_with_leftover_noise}")
if sampler_type == "sdxl":
print(f"start_at_step: {start_at_step}")
if X_type == "Refiner On/Off":
print(f"refine_at_percent: {X_value[0]}")
elif Y_type == "Refiner On/Off":
print(f"refine_at_percent: {Y_value[0]}")
else:
print(f"refine_at_step: {end_at_step}")
print(f"cfg: {cfg}")
if scheduler == "_":
print(f"sampler(scheduler): {sampler_name}")
else:
print(f"sampler: {sampler_name}")
print(f"scheduler: {scheduler}")
if sampler_type == "regular":
print(f"denoise: {denoise}")
if X_type == "Positive Prompt S/R" or Y_type == "Positive Prompt S/R":
positive_prompt = ", ".join([str(x[0]) if i == 0 else str(x[1]) for i, x in enumerate(
X_value)]) if X_type == "Positive Prompt S/R" else ", ".join(
[str(y[0]) if i == 0 else str(y[1]) for i, y in
enumerate(Y_value)]) if Y_type == "Positive Prompt S/R" else positive_prompt
print(f"+prompt_s/r: {positive_prompt}")
if X_type == "Negative Prompt S/R" or Y_type == "Negative Prompt S/R":
negative_prompt = ", ".join([str(x[0]) if i == 0 else str(x[1]) for i, x in enumerate(
X_value)]) if X_type == "Negative Prompt S/R" else ", ".join(
[str(y[0]) if i == 0 else str(y[1]) for i, y in
enumerate(Y_value)]) if Y_type == "Negative Prompt S/R" else negative_prompt
print(f"-prompt_s/r: {negative_prompt}")
if "ControlNet" in X_type or "ControlNet" in Y_type:
cnet_strength, cnet_start_pct, cnet_end_pct = cnet_stack[1]
if "ControlNet" in X_type:
if "Strength" in X_type:
cnet_strength = [str(round(inner_list[0][2], 3)) for inner_list in X_value if
isinstance(inner_list, list) and
inner_list and isinstance(inner_list[0], tuple) and len(inner_list[0]) >= 3]
if "Start%" in X_type:
cnet_start_pct = [str(round(inner_list[0][3], 3)) for inner_list in X_value if
isinstance(inner_list, list) and
inner_list and isinstance(inner_list[0], tuple) and len(inner_list[0]) >= 3]
if "End%" in X_type:
cnet_end_pct = [str(round(inner_list[0][4], 3)) for inner_list in X_value if
isinstance(inner_list, list) and
inner_list and isinstance(inner_list[0], tuple) and len(inner_list[0]) >= 3]
if "ControlNet" in Y_type:
if "Strength" in Y_type:
cnet_strength = [str(round(inner_list[0][2], 3)) for inner_list in Y_value if
isinstance(inner_list, list) and
inner_list and isinstance(inner_list[0], tuple) and len(
inner_list[0]) >= 3]
if "Start%" in Y_type:
cnet_start_pct = [str(round(inner_list[0][3], 3)) for inner_list in Y_value if
isinstance(inner_list, list) and
inner_list and isinstance(inner_list[0], tuple) and len(
inner_list[0]) >= 3]
if "End%" in Y_type:
cnet_end_pct = [str(round(inner_list[0][4], 3)) for inner_list in Y_value if
isinstance(inner_list, list) and
inner_list and isinstance(inner_list[0], tuple) and len(
inner_list[0]) >= 3]
if "ControlNet" in X_type or "ControlNet" in Y_type:
print(f"cnet_strength: {', '.join(cnet_strength) if isinstance(cnet_strength, list) else cnet_strength}")
print(f"cnet_start%: {', '.join(cnet_start_pct) if isinstance(cnet_start_pct, list) else cnet_start_pct}")
print(f"cnet_end%: {', '.join(cnet_end_pct) if isinstance(cnet_end_pct, list) else cnet_end_pct}")
# ______________________________________________________________________________________________________
def adjusted_font_size(text, initial_font_size, i_width):
font = ImageFont.truetype(str(Path(font_path)), initial_font_size)
text_width = font.getlength(text)
if text_width > (i_width * 0.9):
scaling_factor = 0.9 # A value less than 1 to shrink the font size more aggressively
new_font_size = int(initial_font_size * (i_width / text_width) * scaling_factor)
else:
new_font_size = initial_font_size
return new_font_size
# ______________________________________________________________________________________________________
def rearrange_list_A(arr, num_cols, num_rows):
new_list = []
for i in range(num_rows):
for j in range(num_cols):
index = j * num_rows + i
new_list.append(arr[index])
return new_list
def rearrange_list_B(arr, num_rows, num_cols):
new_list = []
for i in range(num_rows):
for j in range(num_cols):
index = i * num_cols + j
new_list.append(arr[index])
return new_list
# Extract plot dimensions
num_rows = max(len(Y_value) if Y_value is not None else 0, 1)
num_cols = max(len(X_value) if X_value is not None else 0, 1)
# Flip X & Y results back if flipped earlier (for Checkpoint/LoRA For loop optimizations)
if flip_xy == True:
X_type, Y_type = Y_type, X_type
X_value, Y_value = Y_value, X_value
X_label, Y_label = Y_label, X_label
num_rows, num_cols = num_cols, num_rows
image_pil_list = rearrange_list_A(image_pil_list, num_rows, num_cols)
else:
image_pil_list = rearrange_list_B(image_pil_list, num_rows, num_cols)
image_tensor_list = rearrange_list_A(image_tensor_list, num_cols, num_rows)
latent_list = rearrange_list_A(latent_list, num_cols, num_rows)
# Extract final image dimensions
i_height, i_width = image_tensor_list[0].shape[1], image_tensor_list[0].shape[2]
# Print XY Plot Results
print_plot_variables(X_type, Y_type, X_value, Y_value, add_noise, seed, steps, start_at_step, end_at_step,
return_with_leftover_noise, cfg, sampler_name, scheduler, denoise, vae_name, ckpt_name,
clip_skip, refiner_name, refiner_clip_skip, ascore, lora_stack, cnet_stack,
sampler_type, num_rows, num_cols, i_height, i_width)
# Concatenate the 'samples' and 'noise_mask' tensors along the first dimension (dim=0)
keys = latent_list[0].keys()
result = {}
for key in keys:
tensors = [d[key] for d in latent_list]
result[key] = torch.cat(tensors, dim=0)
latent_list = result
# Store latent_list as last latent
###update_value_by_id("latent", my_unique_id, latent_list)
# Calculate the dimensions of the white background image
border_size_top = i_width // 15
# Longest Y-label length
if len(Y_label) > 0:
Y_label_longest = max(len(s) for s in Y_label)
else:
# Handle the case when the sequence is empty
Y_label_longest = 0 # or any other appropriate value
Y_label_scale = min(Y_label_longest + 4,24) / 24
if Y_label_orientation == "Vertical":
border_size_left = border_size_top
else: # Assuming Y_label_orientation is "Horizontal"
# border_size_left is now min(i_width, i_height) plus 20% of the difference between the two
border_size_left = min(i_width, i_height) + int(0.2 * abs(i_width - i_height))
border_size_left = int(border_size_left * Y_label_scale)
# Modify the border size, background width and x_offset initialization based on Y_type and Y_label_orientation
if Y_type == "Nothing":
bg_width = num_cols * i_width + (num_cols - 1) * grid_spacing
x_offset_initial = 0
else:
if Y_label_orientation == "Vertical":
bg_width = num_cols * i_width + (num_cols - 1) * grid_spacing + 3 * border_size_left
x_offset_initial = border_size_left * 3
else: # Assuming Y_label_orientation is "Horizontal"
bg_width = num_cols * i_width + (num_cols - 1) * grid_spacing + border_size_left
x_offset_initial = border_size_left
# Modify the background height based on X_type
if X_type == "Nothing":
bg_height = num_rows * i_height + (num_rows - 1) * grid_spacing
y_offset = 0
else:
bg_height = num_rows * i_height + (num_rows - 1) * grid_spacing + 3 * border_size_top
y_offset = border_size_top * 3
# Create the white background image
background = Image.new('RGBA', (int(bg_width), int(bg_height)), color=(255, 255, 255, 255))
for row in range(num_rows):
# Initialize the X_offset
x_offset = x_offset_initial
for col in range(num_cols):
# Calculate the index for image_pil_list
index = col * num_rows + row
img = image_pil_list[index]
# Paste the image
background.paste(img, (x_offset, y_offset))
if row == 0 and X_type != "Nothing":
# Assign text
text = X_label[col]
# Add the corresponding X_value as a label above the image
initial_font_size = int(48 * img.width / 512)
font_size = adjusted_font_size(text, initial_font_size, img.width)
label_height = int(font_size*1.5)
# Create a white background label image
label_bg = Image.new('RGBA', (img.width, label_height), color=(255, 255, 255, 0))
d = ImageDraw.Draw(label_bg)
# Create the font object
font = ImageFont.truetype(str(Path(font_path)), font_size)
# Calculate the text size and the starting position
_, _, text_width, text_height = d.textbbox([0,0], text, font=font)
text_x = (img.width - text_width) // 2
text_y = (label_height - text_height) // 2
# Add the text to the label image
d.text((text_x, text_y), text, fill='black', font=font)
# Calculate the available space between the top of the background and the top of the image
available_space = y_offset - label_height
# Calculate the new Y position for the label image
label_y = available_space // 2
# Paste the label image above the image on the background using alpha_composite()
background.alpha_composite(label_bg, (x_offset, label_y))
if col == 0 and Y_type != "Nothing":
# Assign text
text = Y_label[row]
# Add the corresponding Y_value as a label to the left of the image
if Y_label_orientation == "Vertical":
initial_font_size = int(48 * i_width / 512) # Adjusting this to be same as X_label size
font_size = adjusted_font_size(text, initial_font_size, i_width)
else: # Assuming Y_label_orientation is "Horizontal"
initial_font_size = int(48 * (border_size_left/Y_label_scale) / 512) # Adjusting this to be same as X_label size
font_size = adjusted_font_size(text, initial_font_size, int(border_size_left/Y_label_scale))
# Create a white background label image
label_bg = Image.new('RGBA', (img.height, int(font_size*1.2)), color=(255, 255, 255, 0))
d = ImageDraw.Draw(label_bg)
# Create the font object
font = ImageFont.truetype(str(Path(font_path)), font_size)
# Calculate the text size and the starting position
_, _, text_width, text_height = d.textbbox([0,0], text, font=font)
text_x = (img.height - text_width) // 2
text_y = (font_size - text_height) // 2
# Add the text to the label image
d.text((text_x, text_y), text, fill='black', font=font)
# Rotate the label_bg 90 degrees counter-clockwise only if Y_label_orientation is "Vertical"
if Y_label_orientation == "Vertical":
label_bg = label_bg.rotate(90, expand=True)
# Calculate the available space between the left of the background and the left of the image
available_space = x_offset - label_bg.width
# Calculate the new X position for the label image
label_x = available_space // 2
# Calculate the Y position for the label image based on its orientation
if Y_label_orientation == "Vertical":
label_y = y_offset + (img.height - label_bg.height) // 2
else: # Assuming Y_label_orientation is "Horizontal"
label_y = y_offset + img.height - (img.height - label_bg.height) // 2
# Paste the label image to the left of the image on the background using alpha_composite()
background.alpha_composite(label_bg, (label_x, label_y))
# Update the x_offset
x_offset += img.width + grid_spacing
# Update the y_offset
y_offset += img.height + grid_spacing
xy_plot_image = pil2tensor(background)
# Generate the preview_images
preview_images = PreviewImage().save_images(xy_plot_image)["ui"]["images"]
# Generate output_images
output_images = torch.stack([tensor.squeeze() for tensor in image_tensor_list])
# Set the output_image the same as plot image defined by 'xyplot_as_output_image'
if xyplot_as_output_image == True:
output_images = xy_plot_image
# Print cache if set to true
if cache_models == "True":
print_loaded_objects_entries(xyplot_id, prompt)
print("-" * 40) # Print an empty line followed by a separator line
if sampler_type == "sdxl":
sdxl_tuple = original_model, original_clip, original_positive, original_negative,\
original_refiner_model, original_refiner_clip, original_refiner_positive, original_refiner_negative
result = (sdxl_tuple, latent_list, optional_vae, output_images,)
else:
result = (original_model, original_positive, original_negative, latent_list, optional_vae, output_images,)
return {"ui": {"images": preview_images}, "result": result}