(requested) support AMD 8GB GPUs via Windows DirectML

this update is requested by users

ldm_patched/contrib/external.py

@@ -11,7 +11,7 @@ import math
 import time
 import random
 
-from PIL import Image, ImageOps
+from PIL import Image, ImageOps, ImageSequence
 from PIL.PngImagePlugin import PngInfo
 import numpy as np
 import safetensors.torch
@@ -1412,17 +1412,30 @@ class LoadImage:
     FUNCTION = "load_image"
     def load_image(self, image):
         image_path = ldm_patched.utils.path_utils.get_annotated_filepath(image)
-        i = Image.open(image_path)
-        i = ImageOps.exif_transpose(i)
-        image = i.convert("RGB")
-        image = np.array(image).astype(np.float32) / 255.0
-        image = torch.from_numpy(image)[None,]
-        if 'A' in i.getbands():
-            mask = np.array(i.getchannel('A')).astype(np.float32) / 255.0
-            mask = 1. - torch.from_numpy(mask)
+        img = Image.open(image_path)
+        output_images = []
+        output_masks = []
+        for i in ImageSequence.Iterator(img):
+            i = ImageOps.exif_transpose(i)
+            image = i.convert("RGB")
+            image = np.array(image).astype(np.float32) / 255.0
+            image = torch.from_numpy(image)[None,]
+            if 'A' in i.getbands():
+                mask = np.array(i.getchannel('A')).astype(np.float32) / 255.0
+                mask = 1. - torch.from_numpy(mask)
+            else:
+                mask = torch.zeros((64,64), dtype=torch.float32, device="cpu")
+            output_images.append(image)
+            output_masks.append(mask.unsqueeze(0))
+
+        if len(output_images) > 1:
+            output_image = torch.cat(output_images, dim=0)
+            output_mask = torch.cat(output_masks, dim=0)
         else:
-            mask = torch.zeros((64,64), dtype=torch.float32, device="cpu")
-        return (image, mask.unsqueeze(0))
+            output_image = output_images[0]
+            output_mask = output_masks[0]
+
+        return (output_image, output_mask)
 
     @classmethod
     def IS_CHANGED(s, image):
@@ -1480,13 +1493,10 @@ class LoadImageMask:
         return m.digest().hex()
 
     @classmethod
-    def VALIDATE_INPUTS(s, image, channel):
+    def VALIDATE_INPUTS(s, image):
         if not ldm_patched.utils.path_utils.exists_annotated_filepath(image):
             return "Invalid image file: {}".format(image)
 
-        if channel not in s._color_channels:
-            return "Invalid color channel: {}".format(channel)
-
         return True
 
 class ImageScale:
@@ -1871,6 +1881,7 @@ def init_custom_nodes():
         "nodes_video_model.py",
         "nodes_sag.py",
         "nodes_perpneg.py",
+        "nodes_stable3d.py",
     ]
 
     for node_file in extras_files:

ldm_patched/contrib/external_custom_sampler.py

@@ -89,6 +89,7 @@ class SDTurboScheduler:
         return {"required":
                     {"model": ("MODEL",),
                      "steps": ("INT", {"default": 1, "min": 1, "max": 10}),
+                     "denoise": ("FLOAT", {"default": 1.0, "min": 0, "max": 1.0, "step": 0.01}),
                       }
                }
     RETURN_TYPES = ("SIGMAS",)
@@ -96,8 +97,9 @@ class SDTurboScheduler:
 
     FUNCTION = "get_sigmas"
 
-    def get_sigmas(self, model, steps):
-        timesteps = torch.flip(torch.arange(1, 11) * 100 - 1, (0,))[:steps]
+    def get_sigmas(self, model, steps, denoise):
+        start_step = 10 - int(10 * denoise)
+        timesteps = torch.flip(torch.arange(1, 11) * 100 - 1, (0,))[start_step:start_step + steps]
         sigmas = model.model.model_sampling.sigma(timesteps)
         sigmas = torch.cat([sigmas, sigmas.new_zeros([1])])
         return (sigmas, )

ldm_patched/contrib/external_mask.py

@@ -8,6 +8,7 @@ import ldm_patched.modules.utils
 from ldm_patched.contrib.external import MAX_RESOLUTION
 
 def composite(destination, source, x, y, mask = None, multiplier = 8, resize_source = False):
+    source = source.to(destination.device)
     if resize_source:
         source = torch.nn.functional.interpolate(source, size=(destination.shape[2], destination.shape[3]), mode="bilinear")
 
@@ -22,7 +23,7 @@ def composite(destination, source, x, y, mask = None, multiplier = 8, resize_sou
     if mask is None:
         mask = torch.ones_like(source)
     else:
-        mask = mask.clone()
+        mask = mask.to(destination.device, copy=True)
         mask = torch.nn.functional.interpolate(mask.reshape((-1, 1, mask.shape[-2], mask.shape[-1])), size=(source.shape[2], source.shape[3]), mode="bilinear")
         mask = ldm_patched.modules.utils.repeat_to_batch_size(mask, source.shape[0])
 

ldm_patched/contrib/external_rebatch.py

@@ -101,10 +101,40 @@ class LatentRebatch:
 
         return (output_list,)
 
+class ImageRebatch:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "images": ("IMAGE",),
+                              "batch_size": ("INT", {"default": 1, "min": 1, "max": 4096}),
+                              }}
+    RETURN_TYPES = ("IMAGE",)
+    INPUT_IS_LIST = True
+    OUTPUT_IS_LIST = (True, )
+
+    FUNCTION = "rebatch"
+
+    CATEGORY = "image/batch"
+
+    def rebatch(self, images, batch_size):
+        batch_size = batch_size[0]
+
+        output_list = []
+        all_images = []
+        for img in images:
+            for i in range(img.shape[0]):
+                all_images.append(img[i:i+1])
+
+        for i in range(0, len(all_images), batch_size):
+            output_list.append(torch.cat(all_images[i:i+batch_size], dim=0))
+
+        return (output_list,)
+
 NODE_CLASS_MAPPINGS = {
     "RebatchLatents": LatentRebatch,
+    "RebatchImages": ImageRebatch,
 }
 
 NODE_DISPLAY_NAME_MAPPINGS = {
     "RebatchLatents": "Rebatch Latents",
-}
+    "RebatchImages": "Rebatch Images",
+}

ldm_patched/contrib/external_sag.py

@@ -153,7 +153,7 @@ class SelfAttentionGuidance:
             (sag, _) = ldm_patched.modules.samplers.calc_cond_uncond_batch(model, uncond, None, degraded_noised, sigma, model_options)
             return cfg_result + (degraded - sag) * sag_scale
 
-        m.set_model_sampler_post_cfg_function(post_cfg_function)
+        m.set_model_sampler_post_cfg_function(post_cfg_function, disable_cfg1_optimization=True)
 
         # from diffusers:
         # unet.mid_block.attentions[0].transformer_blocks[0].attn1.patch

ldm_patched/contrib/external_stable3d.py

@@ -0,0 +1,60 @@
+# https://github.com/comfyanonymous/ComfyUI/blob/master/nodes.py 
+
+import torch
+import ldm_patched.contrib.external
+import ldm_patched.modules.utils
+
+def camera_embeddings(elevation, azimuth):
+    elevation = torch.as_tensor([elevation])
+    azimuth = torch.as_tensor([azimuth])
+    embeddings = torch.stack(
+        [
+                torch.deg2rad(
+                    (90 - elevation) - (90)
+                ),  # Zero123 polar is 90-elevation
+                torch.sin(torch.deg2rad(azimuth)),
+                torch.cos(torch.deg2rad(azimuth)),
+                torch.deg2rad(
+                    90 - torch.full_like(elevation, 0)
+                ),
+        ], dim=-1).unsqueeze(1)
+
+    return embeddings
+
+
+class StableZero123_Conditioning:
+    @classmethod
+    def INPUT_TYPES(s):
+        return {"required": { "clip_vision": ("CLIP_VISION",),
+                              "init_image": ("IMAGE",),
+                              "vae": ("VAE",),
+                              "width": ("INT", {"default": 256, "min": 16, "max": ldm_patched.contrib.external.MAX_RESOLUTION, "step": 8}),
+                              "height": ("INT", {"default": 256, "min": 16, "max": ldm_patched.contrib.external.MAX_RESOLUTION, "step": 8}),
+                              "batch_size": ("INT", {"default": 1, "min": 1, "max": 4096}),
+                              "elevation": ("FLOAT", {"default": 0.0, "min": -180.0, "max": 180.0}),
+                              "azimuth": ("FLOAT", {"default": 0.0, "min": -180.0, "max": 180.0}),
+                             }}
+    RETURN_TYPES = ("CONDITIONING", "CONDITIONING", "LATENT")
+    RETURN_NAMES = ("positive", "negative", "latent")
+
+    FUNCTION = "encode"
+
+    CATEGORY = "conditioning/3d_models"
+
+    def encode(self, clip_vision, init_image, vae, width, height, batch_size, elevation, azimuth):
+        output = clip_vision.encode_image(init_image)
+        pooled = output.image_embeds.unsqueeze(0)
+        pixels = ldm_patched.modules.utils.common_upscale(init_image.movedim(-1,1), width, height, "bilinear", "center").movedim(1,-1)
+        encode_pixels = pixels[:,:,:,:3]
+        t = vae.encode(encode_pixels)
+        cam_embeds = camera_embeddings(elevation, azimuth)
+        cond = torch.cat([pooled, cam_embeds.repeat((pooled.shape[0], 1, 1))], dim=-1)
+
+        positive = [[cond, {"concat_latent_image": t}]]
+        negative = [[torch.zeros_like(pooled), {"concat_latent_image": torch.zeros_like(t)}]]
+        latent = torch.zeros([batch_size, 4, height // 8, width // 8])
+        return (positive, negative, {"samples":latent})
+
+NODE_CLASS_MAPPINGS = {
+    "StableZero123_Conditioning": StableZero123_Conditioning,
+}