2.1.844

* maintain clip vision device
* update links in troubleshoot

ldm_patched/contrib/external.py

@@ -1870,6 +1870,7 @@ def init_custom_nodes():
         "nodes_images.py",
         "nodes_video_model.py",
         "nodes_sag.py",
+        "nodes_perpneg.py",
     ]
 
     for node_file in extras_files:

ldm_patched/contrib/external_latent.py

@@ -5,9 +5,7 @@ import torch
 
 def reshape_latent_to(target_shape, latent):
     if latent.shape[1:] != target_shape[1:]:
-        latent.movedim(1, -1)
         latent = ldm_patched.modules.utils.common_upscale(latent, target_shape[3], target_shape[2], "bilinear", "center")
-        latent.movedim(-1, 1)
     return ldm_patched.modules.utils.repeat_to_batch_size(latent, target_shape[0])
 
 
@@ -104,9 +102,32 @@ class LatentInterpolate:
         samples_out["samples"] = st * (m1 * ratio + m2 * (1.0 - ratio))
         return (samples_out,)
 
+class LatentBatch:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "samples1": ("LATENT",), "samples2": ("LATENT",)}}
+
+    RETURN_TYPES = ("LATENT",)
+    FUNCTION = "batch"
+
+    CATEGORY = "latent/batch"
+
+    def batch(self, samples1, samples2):
+        samples_out = samples1.copy()
+        s1 = samples1["samples"]
+        s2 = samples2["samples"]
+
+        if s1.shape[1:] != s2.shape[1:]:
+            s2 = ldm_patched.modules.utils.common_upscale(s2, s1.shape[3], s1.shape[2], "bilinear", "center")
+        s = torch.cat((s1, s2), dim=0)
+        samples_out["samples"] = s
+        samples_out["batch_index"] = samples1.get("batch_index", [x for x in range(0, s1.shape[0])]) + samples2.get("batch_index", [x for x in range(0, s2.shape[0])])
+        return (samples_out,)
+
 NODE_CLASS_MAPPINGS = {
     "LatentAdd": LatentAdd,
     "LatentSubtract": LatentSubtract,
     "LatentMultiply": LatentMultiply,
     "LatentInterpolate": LatentInterpolate,
+    "LatentBatch": LatentBatch,
 }

ldm_patched/contrib/external_model_advanced.py

@@ -19,41 +19,19 @@ class LCM(ldm_patched.modules.model_sampling.EPS):
 
         return c_out * x0 + c_skip * model_input
 
-class ModelSamplingDiscreteDistilled(torch.nn.Module):
+class ModelSamplingDiscreteDistilled(ldm_patched.modules.model_sampling.ModelSamplingDiscrete):
     original_timesteps = 50
 
-    def __init__(self):
-        super().__init__()
-        self.sigma_data = 1.0
-        timesteps = 1000
-        beta_start = 0.00085
-        beta_end = 0.012
+    def __init__(self, model_config=None):
+        super().__init__(model_config)
 
-        betas = torch.linspace(beta_start**0.5, beta_end**0.5, timesteps, dtype=torch.float32) ** 2
-        alphas = 1.0 - betas
-        alphas_cumprod = torch.cumprod(alphas, dim=0)
+        self.skip_steps = self.num_timesteps // self.original_timesteps
 
-        self.skip_steps = timesteps // self.original_timesteps
-
-
-        alphas_cumprod_valid = torch.zeros((self.original_timesteps), dtype=torch.float32)
+        sigmas_valid = torch.zeros((self.original_timesteps), dtype=torch.float32)
         for x in range(self.original_timesteps):
-            alphas_cumprod_valid[self.original_timesteps - 1 - x] = alphas_cumprod[timesteps - 1 - x * self.skip_steps]
+            sigmas_valid[self.original_timesteps - 1 - x] = self.sigmas[self.num_timesteps - 1 - x * self.skip_steps]
 
-        sigmas = ((1 - alphas_cumprod_valid) / alphas_cumprod_valid) ** 0.5
-        self.set_sigmas(sigmas)
-
-    def set_sigmas(self, sigmas):
-        self.register_buffer('sigmas', sigmas)
-        self.register_buffer('log_sigmas', sigmas.log())
-
-    @property
-    def sigma_min(self):
-        return self.sigmas[0]
-
-    @property
-    def sigma_max(self):
-        return self.sigmas[-1]
+        self.set_sigmas(sigmas_valid)
 
     def timestep(self, sigma):
         log_sigma = sigma.log()
@@ -68,14 +46,6 @@ class ModelSamplingDiscreteDistilled(torch.nn.Module):
         log_sigma = (1 - w) * self.log_sigmas[low_idx] + w * self.log_sigmas[high_idx]
         return log_sigma.exp().to(timestep.device)
 
-    def percent_to_sigma(self, percent):
-        if percent <= 0.0:
-            return 999999999.9
-        if percent >= 1.0:
-            return 0.0
-        percent = 1.0 - percent
-        return self.sigma(torch.tensor(percent * 999.0)).item()
-
 
 def rescale_zero_terminal_snr_sigmas(sigmas):
     alphas_cumprod = 1 / ((sigmas * sigmas) + 1)
@@ -124,7 +94,7 @@ class ModelSamplingDiscrete:
         class ModelSamplingAdvanced(sampling_base, sampling_type):
             pass
 
-        model_sampling = ModelSamplingAdvanced()
+        model_sampling = ModelSamplingAdvanced(model.model.model_config)
         if zsnr:
             model_sampling.set_sigmas(rescale_zero_terminal_snr_sigmas(model_sampling.sigmas))
 
@@ -156,7 +126,7 @@ class ModelSamplingContinuousEDM:
         class ModelSamplingAdvanced(ldm_patched.modules.model_sampling.ModelSamplingContinuousEDM, sampling_type):
             pass
 
-        model_sampling = ModelSamplingAdvanced()
+        model_sampling = ModelSamplingAdvanced(model.model.model_config)
         model_sampling.set_sigma_range(sigma_min, sigma_max)
         m.add_object_patch("model_sampling", model_sampling)
         return (m, )

ldm_patched/contrib/external_perpneg.py

@@ -0,0 +1,57 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import torch
+import ldm_patched.modules.model_management
+import ldm_patched.modules.sample
+import ldm_patched.modules.samplers
+import ldm_patched.modules.utils
+
+
+class PerpNeg:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": {"model": ("MODEL", ),
+                             "empty_conditioning": ("CONDITIONING", ),
+                             "neg_scale": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 100.0}),
+                            }}
+    RETURN_TYPES = ("MODEL",)
+    FUNCTION = "patch"
+
+    CATEGORY = "_for_testing"
+
+    def patch(self, model, empty_conditioning, neg_scale):
+        m = model.clone()
+        nocond = ldm_patched.modules.sample.convert_cond(empty_conditioning)
+
+        def cfg_function(args):
+            model = args["model"]
+            noise_pred_pos = args["cond_denoised"]
+            noise_pred_neg = args["uncond_denoised"]
+            cond_scale = args["cond_scale"]
+            x = args["input"]
+            sigma = args["sigma"]
+            model_options = args["model_options"]
+            nocond_processed = ldm_patched.modules.samplers.encode_model_conds(model.extra_conds, nocond, x, x.device, "negative")
+
+            (noise_pred_nocond, _) = ldm_patched.modules.samplers.calc_cond_uncond_batch(model, nocond_processed, None, x, sigma, model_options)
+
+            pos = noise_pred_pos - noise_pred_nocond
+            neg = noise_pred_neg - noise_pred_nocond
+            perp = ((torch.mul(pos, neg).sum())/(torch.norm(neg)**2)) * neg
+            perp_neg = perp * neg_scale
+            cfg_result = noise_pred_nocond + cond_scale*(pos - perp_neg)
+            cfg_result = x - cfg_result
+            return cfg_result
+
+        m.set_model_sampler_cfg_function(cfg_function)
+
+        return (m, )
+
+
+NODE_CLASS_MAPPINGS = {
+    "PerpNeg": PerpNeg,
+}
+
+NODE_DISPLAY_NAME_MAPPINGS = {
+    "PerpNeg": "Perp-Neg",
+}

ldm_patched/contrib/external_sag.py

@@ -60,7 +60,7 @@ def create_blur_map(x0, attn, sigma=3.0, threshold=1.0):
     attn = attn.reshape(b, -1, hw1, hw2)
     # Global Average Pool
     mask = attn.mean(1, keepdim=False).sum(1, keepdim=False) > threshold
-    ratio = round(math.sqrt(lh * lw / hw1))
+    ratio = math.ceil(math.sqrt(lh * lw / hw1))
     mid_shape = [math.ceil(lh / ratio), math.ceil(lw / ratio)]
 
     # Reshape