diff --git a/__init__.py b/__init__.py
index e8fc2b1..32bfecc 100644
--- a/__init__.py
+++ b/__init__.py
@@ -5,7 +5,7 @@ NODE_CLASS_MAPPINGS = {
     "LatentQuilting_Bmad": LatentQuilting,
     "ImageQuiltingSeamlessMB_Bmad": ImageMakeSeamlessMB,
     "ImageQuiltingSeamlessSB_Bmad": ImageMakeSeamlessSB,
-    #"GuessQuiltingBlockSize": GuessBlockSize
+    "GuessQuiltingBlockSize": GuessNiceBlockSize
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
@@ -13,5 +13,5 @@ NODE_DISPLAY_NAME_MAPPINGS = {
     "LatentQuilting_Bmad": "Latent Quilting",
     "ImageQuiltingSeamlessMB_Bmad": "Image Seamless Quilting MP",
     "ImageQuiltingSeamlessSB_Bmad": "Image Seamless Quilting SP",
-    # "GuessQuiltingBlockSize": "Guess BlockSize (Quilting)"
+    "GuessQuiltingBlockSize": "Guess Quilting Block Size"
 }
\ No newline at end of file
diff --git a/guess_block_size.py b/guess_block_size.py
index 24aaed2..cd4e5e4 100644
--- a/guess_block_size.py
+++ b/guess_block_size.py
@@ -27,14 +27,15 @@ def find_sync_wavelen(pairs: size_weight_pairs, lower, upper) -> num_pixels:
     return best_number
 
 
-def make_guess(pairs: size_weight_pairs, min_dim: num_pixels) -> num_pixels:
+def make_guess(pairs: size_weight_pairs, min_dim: num_pixels, max_block_size: num_pixels|None = None) -> num_pixels:
     default_value = round(min_dim / 2.5)  # returned in edge cases
 
     if len(pairs) == 0:  # an edge case; maybe a blank image is sent...
         return default_value
 
     block_size_lower_bound = pairs[0][0]
-    block_size_upper_bound = round(min_dim / 1.2)
+    block_size_upper_bound = round(min_dim / 1.2) if max_block_size is None \
+        else min(round(min_dim / 1.2), max_block_size)
     print(f"initial upper bound = {block_size_upper_bound}")
 
     # the lookup should have at least one single freq sized block of addressable space
@@ -59,7 +60,11 @@ def filter_pairs_by_weight(pairs: size_weight_pairs, weight_percentage_threshold
     return filtered_pairs
 
 
-def guess_nice_block_size(src: np.ndarray) -> num_pixels:
+def guess_nice_block_size(src: np.ndarray, freq_analysis_only: bool = False,
+                          max_block_size: num_pixels|None = None ) -> num_pixels:
+    """
+    @param src: numpy image with normalized float32 values
+    """
     def normalize_weights(pairs: size_weight_pairs):
         if not pairs:
             return []
@@ -71,17 +76,21 @@ def guess_nice_block_size(src: np.ndarray) -> num_pixels:
             normalized_pairs = [(index, 0) for index, weight in pairs]
         return normalized_pairs
 
-    freq_analysis_pairs = analyze_freq_spectrum(src)
-    desc_analysis_pairs = analyze_keypoint_scales(src)
-    # all should come already sorted in descending order w/ respect to weight
+    # src should come with normalized float values already
+    freq_analysis_pairs = analyze_freq_spectrum(src)    # here the image needs to go with float normalized values
+    src = (src*255).astype(np.uint8)
 
+    # here the image needs to go with integer, 0 to 255, values
+    desc_analysis_pairs = [] if freq_analysis_only else analyze_keypoint_scales(src)
+
+    # all pairs should come already sorted in descending order w/ respect to weight
     print(freq_analysis_pairs)
     print(desc_analysis_pairs)
 
     # filter very small distances, with respect to the src size
-    min_dim = min(image.shape[:2])
+    min_dim = min(src.shape[:2])
     thresh_distance = ceil(min_dim ** (1 / 4))
-    block_size_upper_bound = round(min_dim / 1.2)
+    block_size_upper_bound = round(min_dim / 1.2) if max_block_size is None else max_block_size
     freq_analysis_pairs = [(dst, w) for dst, w in freq_analysis_pairs if
                            thresh_distance <= dst < block_size_upper_bound]
     desc_analysis_pairs = [(dst, w) for dst, w in desc_analysis_pairs if
@@ -97,7 +106,7 @@ def guess_nice_block_size(src: np.ndarray) -> num_pixels:
     ]  # may contain duplicates or multiples, that is expected
 
     print(f"final pairs: {final_pairs}")
-    return make_guess(final_pairs, min_dim)
+    return make_guess(final_pairs, min_dim, max_block_size)
 
 
 if __name__ == "__main__":
diff --git a/misc/bse_desc_util.py b/misc/bse_desc_util.py
index 699b00e..54c5a3c 100644
--- a/misc/bse_desc_util.py
+++ b/misc/bse_desc_util.py
@@ -12,6 +12,9 @@ label: TypeAlias = int
 
 
 def find_optimal_clusters(data: list, max_k: int = 6) -> tuple[list[label], list[...], int, float]:
+    if len(np.unique(data)) == 1:  # edge case
+        return [0] * len(data), [data[0]], 1, 0
+
     iters = range(2, max_k + 1)
     best_k = 2
     best_score = -1.0
@@ -81,6 +84,8 @@ def inner_square_area(circle_diameter: float) -> float:
 def analyze_keypoint_scales(image: np.ndarray) -> size_weight_pairs:
     sift = cv2.SIFT_create()
     keypoints = sift.detect(image, None)
+    if len(keypoints) == 0:  # edge case
+        return []
     kp_sizes = [kp.size for kp in keypoints]  # keypoints' diameters, in pixels
     kp_pts = [kp.pt for kp in keypoints]  # keypoints' (y, x) positions
 
@@ -104,7 +109,7 @@ def analyze_keypoint_scales(image: np.ndarray) -> size_weight_pairs:
 
 
 if __name__ == "__main__":
-    image_path = '../t18.png'
+    image_path = '../t9.png'
     image = cv2.imread(image_path)
     data = analyze_keypoint_scales(image)
     print(data)
diff --git a/misc/bse_ft_util.py b/misc/bse_ft_util.py
index f98bebc..d581166 100644
--- a/misc/bse_ft_util.py
+++ b/misc/bse_ft_util.py
@@ -4,8 +4,8 @@ import cv2
 
 
 def compute_fft(image):
-    image_float = np.float32(image)
-    dft = cv2.dft(image_float, flags=cv2.DFT_COMPLEX_OUTPUT)
+    image = np.float32(image)
+    dft = cv2.dft(image, flags=cv2.DFT_COMPLEX_OUTPUT)
     dft_shift = np.fft.fftshift(dft)
     magnitude_spectrum = cv2.magnitude(dft_shift[:, :, 0], dft_shift[:, :, 1])
     return magnitude_spectrum
diff --git a/nodes.py b/nodes.py
index 7a17186..e6a41bc 100644
--- a/nodes.py
+++ b/nodes.py
@@ -16,38 +16,42 @@ import cv2
 NODES_CATEGORY = "Bmad/CV/Quilting"
 SEAMLESS_DIRS = ["H", "V", "H & V"]  # options for seamless nodes
 
-QUILTING_SHARED_INPUT_TYPES = {
-    # block size is given in pixels
-    "block_size": ("INT", {"default": 20, "min": 3, "max": 256, "step": 1}),
 
-    # the percentage of pixels that overlap between each block_sized block
-    "overlap": ("FLOAT", {"default": 1 / 6.0, "min": .1, "max": .9, "step": .01}),
+def get_quilting_shared_input_types():
+    return {
+        # block size is given in pixels.
+        # if inferior to 3, guess_nice_block_size is used instead.
+        # if negative, guess_nice_block_size only does freq. analysis (should be considerably faster)
+        "block_size": ("INT", {"default": -1, "min": -1, "max": 512, "step": 1}),
 
-    # this is a percentage relative to min error when searching for a patch.
-    # the ones that within tolerance are potential candidates to be selected.
-    # tolerance equal to 1 means a tolerance of 2 times the min error.
-    # my interpretation regarding its application is that
-    # tolerance can help prevent too much sameness in the texture
-    # due to some subset of patches being better at minimizing the error.
-    "tolerance": ("FLOAT", {"default": .1, "min": 0, "max": 2, "step": .01}),
+        # the percentage of pixels that overlap between each block_sized block
+        "overlap": ("FLOAT", {"default": 1 / 5.0, "min": .1, "max": .9, "step": .01}),
 
-    # 0 -> no parallelization; uses the reference implementation
-    # 1 -> 4 jobs used ( divides generation into 4 sections )
-    # 2 and above ->  the number of jobs per section
-    # parallelization lvl also affects output; even if seed is fixed, changing this will change the output
-    "parallelization_lvl": ("INT", {"default": 1, "min": 0, "max": 6, "step": 1}),
+        # this is a percentage relative to min error when searching for a patch.
+        # the ones that within tolerance are potential candidates to be selected.
+        # tolerance equal to 1 means a tolerance of 2 times the min error.
+        # my interpretation regarding its application is that
+        # tolerance can help prevent too much sameness in the texture
+        # due to some subset of patches being better at minimizing the error.
+        "tolerance": ("FLOAT", {"default": .1, "min": 0, "max": 2, "step": .01}),
 
-    "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
+        # 0 -> no parallelization; uses the reference implementation
+        # 1 -> 4 jobs used ( divides generation into 4 sections )
+        # 2 and above ->  the number of jobs per section
+        # parallelization lvl also affects output; even if seed is fixed, changing this will change the output
+        "parallelization_lvl": ("INT", {"default": 1, "min": 0, "max": 6, "step": 1}),
 
-    "version": ("INT", {"default": 1, "min": 0, "max": 3}),
-}
+        "seed": ("INT", {"default": 0, "min": 0, "max": 0xffffffffffffffff}),
+
+        "version": ("INT", {"default": 1, "min": 0, "max": 3}),
+    }
 
 
 @dataclass
 class QuiltingFuncWrapper:
     """Wraps node functionality for easy re-use when using jobs."""
-    block_size: int
-    overlap: int
+    block_sizes: list[int]
+    overlap: float
     out_h: int
     out_w: int
     tolerance: float
@@ -94,29 +98,43 @@ def terminate_generation(finished_event, jobs_shared_memory, pbt: Thread):
         throw_exception_if_processing_interrupted()
 
 
-def setup_pbar_quilting(block_size, overlap, out_height, out_width, par_lvl, batch_len):
-    n_rows = int(np.ceil((out_height - block_size) * 1.0 / (block_size - overlap)))
-    n_columns = int(np.ceil((out_width - block_size) * 1.0 / (block_size - overlap)))
-    total_steps: int = ((n_rows + 1) * (n_columns + 1) - 1) * batch_len  # ignores first corner/center
+def setup_pbar_quilting(block_sizes: list[int], overlap_percentage: float, out_height, out_width, par_lvl):
+    """
+    @param block_sizes: block size for each item in the batch. len(block_sizes) = batch length
+    """
+    total_steps: int = 0
+    batch_len: int = len(block_sizes)
+
+    for block_size in block_sizes:
+        overlap = overlap_percentage_to_pixels(block_size, overlap_percentage)
+        n_rows = int(np.ceil((out_height - block_size) * 1.0 / (block_size - overlap)))
+        n_columns = int(np.ceil((out_width - block_size) * 1.0 / (block_size - overlap)))
+        total_steps += ((n_rows + 1) * (n_columns + 1) - 1)  # ignores first corner/center
     # might be less than the generated when using parallel solution, but not by far, so will leave it like this for now
 
     return setup_pbar(total_steps, par_lvl, batch_len)
 
 
-def setup_pbar_seamless(ori, block_size, overlap, height, width, batch_len):
+def setup_pbar_seamless(ori, block_sizes: list[int], overlap_percentage: float, height, width, batch_len):
+    """
+    here batch_len may differ from block_sizes due to the number of lookup textures
+    """
     from .make_seamless import get_numb_of_blocks_to_fill_stripe
-    match ori:
-        case "H":
-            total_steps = get_numb_of_blocks_to_fill_stripe(block_size, overlap, width)
-        case "V":
-            total_steps = get_numb_of_blocks_to_fill_stripe(block_size, overlap, height)
-        case _:
-            total_steps = (
-                    2 +
-                    get_numb_of_blocks_to_fill_stripe(block_size, overlap, height) +
-                    get_numb_of_blocks_to_fill_stripe(block_size, overlap, width)
-            )
-    total_steps = total_steps * batch_len
+    total_steps: int = 0
+    for i in range(batch_len):
+        block_size = block_sizes[min(i, len(block_sizes) - 1)]
+        overlap = overlap_percentage_to_pixels(block_size, overlap_percentage)
+        match ori:
+            case "H":
+                total_steps += get_numb_of_blocks_to_fill_stripe(block_size, overlap, width)
+            case "V":
+                total_steps += get_numb_of_blocks_to_fill_stripe(block_size, overlap, height)
+            case _:
+                total_steps += (
+                        2 +
+                        get_numb_of_blocks_to_fill_stripe(block_size, overlap, height) +
+                        get_numb_of_blocks_to_fill_stripe(block_size, overlap, width)
+                )
     return setup_pbar(total_steps, 0, batch_len)
 
 
@@ -124,7 +142,7 @@ def setup_pbar_seamless_v2(ori, batch_len):
     # 3 increments per big block + 2 for the "H & V" H Seam patch
     match ori:
         case "H & V":
-            total_steps = 2 + 3*2
+            total_steps = 2 + 3 * 2
         case _:
             total_steps = 3
     total_steps = total_steps * batch_len
@@ -154,6 +172,14 @@ def setup_pbar(total_steps, par_lvl, batch_len):
     return finished_event, t, shm_jobs.name, shm_jobs
 
 
+def unwrap_to_grey(image, is_latent: bool = False) -> np.ndarray:
+    squeeze = len(image.shape) > 3
+    image = image.cpu().numpy()
+    image = image.squeeze() if squeeze else image
+    image = np.moveaxis(image, 0, -1) if is_latent else image
+    return cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
+
+
 def unwrap_and_quilt(wrapped_func, image, job_id, is_latent: bool = False):
     """quilting job when using batches"""
     squeeze = len(image.shape) > 3
@@ -181,9 +207,40 @@ def unwrap_and_quilt_seamless(wrapped_func, image, lookup, job_id):
     return result
 
 
-def overlap_percentage_to_pixels(block_size, overlap):
+def overlap_percentage_to_pixels(block_size: int, overlap: float):
     return int(block_size * overlap) if overlap > 0 else int(
-        block_size * QUILTING_SHARED_INPUT_TYPES["overlap"][1]["default"])
+        block_size * get_quilting_shared_input_types()["overlap"][1]["default"])
+
+
+def get_block_sizes(src, block_size_input: int, block_size_upper_bound: int|None = None):
+    if block_size_input >= 3:
+        return [block_size_input] * src.shape[0]
+
+    print(f"block size set to {block_size_input} less than 3!\nguessing nice block size...")
+    only_do_freq_analysis = block_size_input < 0
+    sizes = [guess_nice_block_size(unwrap_to_grey(src[i]), only_do_freq_analysis, block_size_upper_bound)
+             for i in range(src.shape[0])]
+    print(f"guessed block sizes: {sizes}")
+
+    return sizes
+
+
+def block_size_upper_bound_for_seamless(ori, tex_h, tex_w, overlap_percentage):
+    """
+    if using guess_block_size in seamless nodes, H seam
+    """
+    match ori:
+        case "H & V":
+            # mind H seam w/ the following related assert -> image.shape[1] >= block_size + overlap * 2
+            # -> H >= S + S * OP * 2 <-> H / (1 + 2 * OP ) >= S
+            return min(
+                # similar value to the one computed in guess nice block
+                round(min(tex_w, tex_h) / 1.2),
+                # space required to patch h seam  plus some extra margin
+                round(tex_h/(1+2*overlap_percentage)/1.1)
+            )
+        case _______:
+            return None  # texture default dims are fine as bounds
 
 # endregion AUX FUNCTIONS & CLASSES
 
@@ -198,13 +255,16 @@ class ImageQuilting:
             if self.version == 2:
                 image = cv2.cvtColor(image, cv2.COLOR_RGB2Lab)
 
+            block_size = self.block_sizes[job_id]
+            overlap = overlap_percentage_to_pixels(block_size, self.overlap)
+
             if self.parallelization_lvl == 0:
                 result = generate_texture(
-                    image, self.block_size, self.overlap, self.out_h, self.out_w, self.tolerance,
+                    image, block_size, overlap, self.out_h, self.out_w, self.tolerance,
                     self.version, self.rng, UiCoordData(self.jobs_shm_name, job_id))
             else:
                 result = generate_texture_parallel(
-                    image, self.block_size, self.overlap, self.out_h, self.out_w, self.tolerance,
+                    image, block_size, overlap, self.out_h, self.out_w, self.tolerance,
                     self.version, self.parallelization_lvl, self.rng, UiCoordData(self.jobs_shm_name, job_id))
 
             if self.version == 2:
@@ -223,7 +283,7 @@ class ImageQuilting:
                 # self explanatory
                 "scale": ("FLOAT", {"default": 4, "min": 2, "max": 10, "step": .1}),
 
-                **QUILTING_SHARED_INPUT_TYPES
+                **get_quilting_shared_input_types()
             }
         }
 
@@ -234,17 +294,17 @@ class ImageQuilting:
     def compute(self, src, block_size, scale, overlap, tolerance, parallelization_lvl, seed, version):
         h, w = src.shape[1:3]
         out_h, out_w = int(scale * h), int(scale * w)
-        overlap = overlap_percentage_to_pixels(block_size, overlap)
+        block_sizes = get_block_sizes(src, block_size)
 
         # note: the input src should have normalized values, not 0 to 255
 
         rng: numpy.random.Generator = np.random.default_rng(seed=seed)
 
         finish_event, t, shm_name, shm_jobs = \
-            setup_pbar_quilting(block_size, overlap, out_h, out_w, parallelization_lvl, src.shape[0])
+            setup_pbar_quilting(block_sizes, overlap, out_h, out_w, parallelization_lvl)
 
         func = ImageQuilting.ImageQuiltingFuncWrapper(
-            block_size, overlap, out_h, out_w, tolerance, parallelization_lvl, rng, version, shm_name)
+            block_sizes, overlap, out_h, out_w, tolerance, parallelization_lvl, rng, version, shm_name)
 
         if src.shape[0] > 1:  # if image batch
             texture_batch = self.batch_using_jobs(func, src)
@@ -269,17 +329,25 @@ class LatentQuilting:
         """Wraps node functionality for easy re-use when using jobs."""
 
         def __call__(self, latent_image, job_id):
+            block_size = self.block_sizes[job_id]
+            overlap = overlap_percentage_to_pixels(block_size, self.overlap)
             if self.parallelization_lvl == 0:
                 return generate_texture(
-                    latent_image, self.block_size, self.overlap, self.out_h, self.out_w, self.tolerance,
+                    latent_image, block_size, overlap, self.out_h, self.out_w, self.tolerance,
                     self.version, self.rng, UiCoordData(self.jobs_shm_name, job_id))
             else:
                 return generate_texture_parallel(
-                    latent_image, self.block_size, self.overlap, self.out_h, self.out_w, self.tolerance,
+                    latent_image, block_size, overlap, self.out_h, self.out_w, self.tolerance,
                     self.version, self.parallelization_lvl, self.rng, UiCoordData(self.jobs_shm_name, job_id))
 
     @classmethod
     def INPUT_TYPES(cls):
+        shared_inputs = get_quilting_shared_input_types()
+        # don't allow for auto block size
+        shared_inputs["block_size"][1]["default"] = 18
+        shared_inputs["block_size"][1]["min"] = 3
+        # note: could try running over individual channels; not sure if worth it thought.
+
         return {
             "required": {
                 "src": ("LATENT",),
@@ -287,7 +355,7 @@ class LatentQuilting:
                 # self explanatory
                 "scale": ("FLOAT", {"default": 4, "min": 2, "max": 32, "step": .1}),
 
-                **QUILTING_SHARED_INPUT_TYPES
+                **shared_inputs
             }
         }
 
@@ -299,15 +367,15 @@ class LatentQuilting:
         src = src["samples"]
         h, w = src.shape[2:4]
         out_h, out_w = int(scale * h), int(scale * w)
-        overlap = overlap_percentage_to_pixels(block_size, overlap)
+        block_sizes = [block_size] * src.shape[0]
 
         rng: numpy.random.Generator = np.random.default_rng(seed=seed)
 
         finish_event, t, shm_name, shm_jobs = \
-            setup_pbar_quilting(block_size, overlap, out_h, out_w, parallelization_lvl, src.shape[0])
+            setup_pbar_quilting(block_sizes, overlap, out_h, out_w, parallelization_lvl)
 
         func = LatentQuilting.LatentQuiltingFuncWrapper(
-            block_size, overlap, out_h, out_w, tolerance, parallelization_lvl, rng, version, shm_name)
+            block_sizes, overlap, out_h, out_w, tolerance, parallelization_lvl, rng, version, shm_name)
 
         if src.shape[0] > 1:  # if multiple
             latent_batch = self.batch_using_jobs(func, src)
@@ -335,7 +403,7 @@ class ImageMakeSeamlessMB:
         """Wraps node functionality for easy re-use when using jobs."""
 
         ori: str
-        block_size: int
+        block_sizes: list[int]
         overlap: int
         tolerance: float
         rng: numpy.random.Generator
@@ -345,6 +413,9 @@ class ImageMakeSeamlessMB:
         def __call__(self, image, lookup, job_id):
             from .make_seamless import make_seamless_horizontally, make_seamless_vertically, make_seamless_both
 
+            block_size = self.block_sizes[min(job_id, len(self.block_sizes) - 1)]
+            overlap = overlap_percentage_to_pixels(block_size, self.overlap)
+
             if self.version == 2:
                 image = cv2.cvtColor(image, cv2.COLOR_RGB2Lab)
                 if lookup is not None:
@@ -358,7 +429,7 @@ class ImageMakeSeamlessMB:
                 case ___:
                     func = make_seamless_both
 
-            result = func(image, self.block_size, self.overlap, self.tolerance,
+            result = func(image, block_size, overlap, self.tolerance,
                           self.rng, self.version, lookup, UiCoordData(self.jobs_shm_name, job_id))
 
             if self.version == 2:
@@ -368,7 +439,7 @@ class ImageMakeSeamlessMB:
 
     @classmethod
     def INPUT_TYPES(cls):
-        inputs = QUILTING_SHARED_INPUT_TYPES.copy()
+        inputs = get_quilting_shared_input_types()
         inputs.pop("parallelization_lvl")
         inputs["version"][1]["min"] = 1
         inputs["overlap"][1]["max"] = .5
@@ -391,15 +462,16 @@ class ImageMakeSeamlessMB:
         # note that src = lookup is the current algorithm policy when lookup is not provided.
         # this policy could change in the future, so do not apply it here too despite being idempotent.
         h, w = src.shape[1:3]
-        overlap = overlap_percentage_to_pixels(block_size, overlap)
+        blk_size_upper_bound = block_size_upper_bound_for_seamless(ori, h, w, overlap)
+        block_sizes = get_block_sizes(src, block_size, blk_size_upper_bound)
         rng: numpy.random.Generator = np.random.default_rng(seed=seed)
 
         lookup_batch_size = lookup.shape[0] if lookup is not None else 0
         finish_event, t, shm_name, shm_jobs = setup_pbar_seamless(
-            ori, block_size, overlap, h, w, max(src.shape[0], lookup_batch_size))
+            ori, block_sizes, overlap, h, w, max(src.shape[0], lookup_batch_size))
 
         func = ImageMakeSeamlessMB.SeamlessFuncWrapper(
-            ori, block_size, overlap, tolerance, rng, version, shm_name)
+            ori, block_sizes, overlap, tolerance, rng, version, shm_name)
 
         if src.shape[0] > 1 or lookup_batch_size > 1:  # if image batch
             texture_batch = self.batch_using_jobs(func, src, lookup)
@@ -415,13 +487,14 @@ class ImageMakeSeamlessMB:
     def batch_using_jobs(wrapped_func, src, lookup):
         from joblib import Parallel, delayed
         # process in the same fashion as lists
+        number_of_processes = src.shape[0] if lookup is None else max(src.shape[0], lookup.shape[0])
         results = Parallel(n_jobs=-1, backend="loky", timeout=None)(
             delayed(unwrap_and_quilt_seamless)(
                 wrapped_func,
                 src[min(i, src.shape[0] - 1)],
                 lookup[min(i, lookup.shape[0] - 1)]
                 if lookup is not None else None,
-                i) for i in range(max(src.shape[0], lookup.shape[0])))
+                i) for i in range(number_of_processes))
         return torch.stack(results)
 
 
@@ -433,7 +506,7 @@ class ImageMakeSeamlessSB:
         """Wraps node functionality for easy re-use when using jobs."""
 
         ori: str
-        block_size: int
+        block_sizes: list[int]
         overlap: int
         rng: numpy.random.Generator
         version: int
@@ -442,6 +515,9 @@ class ImageMakeSeamlessSB:
         def __call__(self, image, lookup, job_id):
             from .make_seamless2 import seamless_horizontal, seamless_vertical, seamless_both
 
+            block_size = self.block_sizes[min(job_id, len(self.block_sizes) - 1)]
+            overlap = overlap_percentage_to_pixels(block_size, self.overlap)
+
             if self.version == 2:
                 image = cv2.cvtColor(image, cv2.COLOR_RGB2Lab)
                 if lookup is not None:
@@ -455,7 +531,7 @@ class ImageMakeSeamlessSB:
                 case ___:
                     func = seamless_both
 
-            result = func(image, self.block_size, self.overlap,
+            result = func(image, block_size, overlap,
                           self.version, lookup, self.rng, UiCoordData(self.jobs_shm_name, job_id))
 
             if self.version == 2:
@@ -465,7 +541,7 @@ class ImageMakeSeamlessSB:
 
     @classmethod
     def INPUT_TYPES(cls):
-        inputs = QUILTING_SHARED_INPUT_TYPES.copy()
+        inputs = get_quilting_shared_input_types()
         inputs.pop("parallelization_lvl")
         inputs.pop("tolerance")
         inputs["version"][1]["min"] = 1
@@ -488,14 +564,16 @@ class ImageMakeSeamlessSB:
     def compute(self, src, ori, block_size, overlap, seed, version, lookup=None):
         # note that src = lookup is the current algorithm policy when lookup is not provided.
         # this policy could change in the future, so do not apply it here too despite being idempotent.
-        overlap = overlap_percentage_to_pixels(block_size, overlap)
+        h, w = src.shape[1:3]
+        blk_size_upper_bound = block_size_upper_bound_for_seamless(ori, h, w, overlap)
+        block_sizes = get_block_sizes(src, block_size, blk_size_upper_bound)
         rng: numpy.random.Generator = np.random.default_rng(seed=seed)
 
         lookup_batch_size = lookup.shape[0] if lookup is not None else 0
         finish_event, t, shm_name, shm_jobs = setup_pbar_seamless_v2(ori, max(src.shape[0], lookup_batch_size))
 
         func = ImageMakeSeamlessSB.SeamlessFuncWrapper(
-            ori, block_size, overlap, rng, version, shm_name)
+            ori, block_sizes, overlap, rng, version, shm_name)
 
         if src.shape[0] > 1 or lookup is not None and lookup.shape[0] > 1:  # if image batch
             texture_batch = self.batch_using_jobs(func, src, lookup)
@@ -511,13 +589,32 @@ class ImageMakeSeamlessSB:
     def batch_using_jobs(wrapped_func, src, lookup):
         from joblib import Parallel, delayed
         # process in the same fashion as lists
+        number_of_processes = src.shape[0] if lookup is None else max(src.shape[0], lookup.shape[0])
         results = Parallel(n_jobs=-1, backend="loky", timeout=None)(
             delayed(unwrap_and_quilt_seamless)(
                 wrapped_func,
                 src[min(i, src.shape[0] - 1)],
                 lookup[min(i, lookup.shape[0] - 1)]
                 if lookup is not None else None,
-                i) for i in range(max(src.shape[0], lookup.shape[0])))
+                i) for i in range(number_of_processes))
         return torch.stack(results)
 
+
+class GuessNiceBlockSize:
+    @classmethod
+    def INPUT_TYPES(cls):
+        return {
+            "required": {
+                "src": ("IMAGE",),
+                "simple_and_fast": ("BOOLEAN", {"default": False})
+            }
+        }
+
+    RETURN_TYPES = ("INT",)
+    FUNCTION = "compute"
+    CATEGORY = NODES_CATEGORY
+
+    def compute(self, src, simple_and_fast):
+        return (guess_nice_block_size(unwrap_to_grey(src[0]), freq_analysis_only=simple_and_fast),)
+
 # endregion NODES