improve: FloatRange - support decrement

Add cosine interpolation for Scheduled CFG

__init__.py

@@ -7,7 +7,7 @@
 
 import importlib
 
-version_code = [0, 82, 5]
+version_code = [0, 84]
 version_str = f"V{version_code[0]}.{version_code[1]}" + (f'.{version_code[2]}' if len(version_code) > 2 else '')
 print(f"### Loading: ComfyUI-Inspire-Pack ({version_str})")
 

inspire/list_nodes.py

@@ -18,9 +18,14 @@ class FloatRange:
     CATEGORY = "InspirePack/Util"
 
     def doit(self, start, stop, step, limit, ensure_end):
-        if start >= stop or step == 0:
+        if start == stop or step == 0:
             return ([start], )
 
+        reverse = False
+        if start > stop:
+            reverse = True
+            start, stop = stop, start
+
         res = []
         x = start
         last = x
@@ -36,6 +41,9 @@ class FloatRange:
 
             res.append(stop)
 
+        if reverse:
+            res.reverse()
+
         return (res, )
 
 

inspire/prompt_support.py

@@ -121,7 +121,7 @@ class LoadPromptsFromFile:
 
         prompts = []
         try:
-            if text_data_opt is None:
+            if not text_data_opt:
                 with open(prompt_path, "r", encoding="utf-8") as file:
                     prompt_data = file.read()
             else:
@@ -182,7 +182,7 @@ class LoadSinglePromptFromFile:
 
         prompts = []
         try:
-            if text_data_opt is None:
+            if not text_data_opt:
                 with open(prompt_path, "r", encoding="utf-8") as file:
                     prompt_data = file.read()
             else:

inspire/sampler_nodes.py

@@ -24,6 +24,7 @@ class KSampler_progress(a1111_compat.KSampler_inspire):
                     "noise_mode": (["GPU(=A1111)", "CPU"],),
                     "interval": ("INT", {"default": 1, "min": 1, "max": 10000}),
                     "omit_start_latent": ("BOOLEAN", {"default": True, "label_on": "True", "label_off": "False"}),
+                    "omit_final_latent": ("BOOLEAN", {"default": False, "label_on": "True", "label_off": "False"}),
                     },
                 "optional": {
                     "scheduler_func_opt": ("SCHEDULER_FUNC",),
@@ -36,7 +37,8 @@ class KSampler_progress(a1111_compat.KSampler_inspire):
     RETURN_NAMES = ("latent", "progress_latent")
 
     @staticmethod
-    def doit(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise, noise_mode, interval, omit_start_latent, scheduler_func_opt=None):
+    def doit(model, seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, denoise, noise_mode,
+             interval, omit_start_latent, omit_final_latent, scheduler_func_opt=None):
         adv_steps = int(steps / denoise)
 
         if omit_start_latent:
@@ -44,19 +46,20 @@ class KSampler_progress(a1111_compat.KSampler_inspire):
         else:
             result = [latent_image['samples']]
 
-        result = []
-
         def progress_callback(step, x0, x, total_steps):
             if (total_steps-1) != step and step % interval != 0:
                 return
 
             x = model.model.process_latent_out(x)
-            x = x.to(model_management.intermediate_device())
+            x = x.cpu()
             result.append(x)
 
         latent_image, noise = a1111_compat.KSamplerAdvanced_inspire.sample(model, True, seed, adv_steps, cfg, sampler_name, scheduler, positive, negative, latent_image, (adv_steps-steps),
                                                                            adv_steps, noise_mode, False, callback=progress_callback, scheduler_func_opt=scheduler_func_opt)
 
+        if not omit_final_latent:
+            result.append(latent_image['samples'].cpu())
+
         if len(result) > 0:
             result = torch.cat(result)
             result = {'samples': result}
@@ -86,6 +89,7 @@ class KSamplerAdvanced_progress(a1111_compat.KSamplerAdvanced_inspire):
                     "return_with_leftover_noise": ("BOOLEAN", {"default": False, "label_on": "enable", "label_off": "disable"}),
                     "interval": ("INT", {"default": 1, "min": 1, "max": 10000}),
                     "omit_start_latent": ("BOOLEAN", {"default": False, "label_on": "True", "label_off": "False"}),
+                    "omit_final_latent": ("BOOLEAN", {"default": False, "label_on": "True", "label_off": "False"}),
                     },
                 "optional": {
                     "prev_progress_latent_opt": ("LATENT",),
@@ -100,26 +104,29 @@ class KSamplerAdvanced_progress(a1111_compat.KSamplerAdvanced_inspire):
     RETURN_TYPES = ("LATENT", "LATENT")
     RETURN_NAMES = ("latent", "progress_latent")
 
-    def doit(self, model, add_noise, noise_seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, start_at_step, end_at_step,
-             noise_mode, return_with_leftover_noise, interval, omit_start_latent, prev_progress_latent_opt=None, scheduler_func_opt=None):
+    def doit(self, model, add_noise, noise_seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image,
+             start_at_step, end_at_step, noise_mode, return_with_leftover_noise, interval, omit_start_latent, omit_final_latent,
+             prev_progress_latent_opt=None, scheduler_func_opt=None):
+
         if omit_start_latent:
             result = []
         else:
             result = [latent_image['samples']]
 
-        result = []
-
         def progress_callback(step, x0, x, total_steps):
             if (total_steps-1) != step and step % interval != 0:
                 return
 
             x = model.model.process_latent_out(x)
-            x = x.to(model_management.intermediate_device())
+            x = x.cpu()
             result.append(x)
 
         latent_image, noise = a1111_compat.KSamplerAdvanced_inspire.sample(model, add_noise, noise_seed, steps, cfg, sampler_name, scheduler, positive, negative, latent_image, start_at_step, end_at_step,
                                                                            noise_mode, False, callback=progress_callback, scheduler_func_opt=scheduler_func_opt)
 
+        if not omit_final_latent:
+            result.append(latent_image['samples'].cpu())
+
         if len(result) > 0:
             result = torch.cat(result)
             result = {'samples': result}
@@ -165,6 +172,15 @@ def logarithmic_interpolation(from_cfg, to_cfg, i, steps):
     return from_cfg + (to_cfg - from_cfg) * t
 
 
+def cosine_interpolation(from_cfg, to_cfg, i, steps):
+    if (i == 0) or (i == steps-1):
+        return from_cfg
+
+    t = (1.0 + math.cos(math.pi*2*(i/steps))) / 2
+
+    return from_cfg + (to_cfg - from_cfg) * t
+
+
 class Guider_scheduled(CFGGuider):
     def __init__(self, model_patcher, sigmas, from_cfg, to_cfg, schedule):
         super().__init__(model_patcher)
@@ -194,6 +210,8 @@ class Guider_scheduled(CFGGuider):
                 self.cfg_sigmas[k] = exponential_interpolation(self.from_cfg, self.to_cfg, i, steps), i
             elif self.schedule == 'log':
                 self.cfg_sigmas[k] = logarithmic_interpolation(self.from_cfg, self.to_cfg, i, steps), i
+            elif self.schedule == 'cos':
+                self.cfg_sigmas[k] = cosine_interpolation(self.from_cfg, self.to_cfg, i, steps), i
             else:
                 self.cfg_sigmas[k] = self.from_cfg + delta * i / steps, i
 
@@ -245,6 +263,8 @@ class Guider_PerpNeg_scheduled(Guider_PerpNeg):
                 self.cfg_sigmas[k] = exponential_interpolation(self.from_cfg, self.to_cfg, i, steps), i
             elif self.schedule == 'log':
                 self.cfg_sigmas[k] = logarithmic_interpolation(self.from_cfg, self.to_cfg, i, steps), i
+            elif self.schedule == 'cos':
+                self.cfg_sigmas[k] = cosine_interpolation(self.from_cfg, self.to_cfg, i, steps), i
             else:
                 self.cfg_sigmas[k] = self.from_cfg + delta * i / steps, i
 
@@ -276,7 +296,7 @@ class ScheduledCFGGuider:
                     "sigmas": ("SIGMAS", ),
                     "from_cfg": ("FLOAT", {"default": 6.5, "min": 0.0, "max": 100.0, "step": 0.1, "round": 0.01}),
                     "to_cfg": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.1, "round": 0.01}),
-                    "schedule": (["linear", "log", "exp"], {'default': 'log'})
+                    "schedule": (["linear", "log", "exp", "cos"], {'default': 'log'})
                     }
                 }
 
@@ -303,7 +323,7 @@ class ScheduledPerpNegCFGGuider:
                     "sigmas": ("SIGMAS", ),
                     "from_cfg": ("FLOAT", {"default": 6.5, "min": 0.0, "max": 100.0, "step": 0.1, "round": 0.01}),
                     "to_cfg": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0, "step": 0.1, "round": 0.01}),
-                    "schedule": (["linear", "log", "exp"], {'default': 'log'})
+                    "schedule": (["linear", "log", "exp", "cos"], {'default': 'log'})
                     }
                 }
 

pyproject.toml

@@ -1,8 +1,8 @@
 [project]
 name = "comfyui-inspire-pack"
 description = "This extension provides various nodes to support Lora Block Weight and the Impact Pack. Provides many easily applicable regional features and applications for Variation Seed."
-version = "0.82.5"
-license = "LICENSE"
+version = "0.84"
+license = { file = "LICENSE" }
 dependencies = ["matplotlib", "cachetools"]
 
 [project.urls]