feat: layer diffuse

hordelib/horde.py

@@ -174,6 +174,7 @@ class HordeLib:
         "extra_source_images": {"datatype": list, "default": []},  # Stable Cascade Remix
         "extra_texts": {"datatype": list, "default": []},  # QR Codes (for now)
         "workflow": {"datatype": str, "default": "auto_detect"},
+        "transparent": {"datatype": bool, "default": False},
     }
 
     EXTRA_IMAGES_SCHEMA = {
@@ -1029,6 +1030,22 @@ def _final_pipeline_adjustments(self, payload, pipeline_data) -> tuple[dict, lis
                     pipeline_params["controlnet_qr_model_loader.control_net_name"] = (
                         "control_v1p_sd15_qrcode_monster_v2.safetensors"
                     )
+        if payload.get("transparent") is True:
+            if SharedModelManager.manager.compvis:
+                model_details = SharedModelManager.manager.compvis.get_model_reference_info(payload["model_name"])
+                # SD2, Cascade and SD3 not supported
+                if model_details and model_details.get("baseline") in ["stable diffusion 1", "stable_diffusion_xl"]:
+                    self.generator.reconnect_input(pipeline_data, "sampler.model", "layer_diffuse_apply")
+                    self.generator.reconnect_input(pipeline_data, "layer_diffuse_apply.model", "model_loader")
+                    self.generator.reconnect_input(pipeline_data, "output_image.images", "layer_diffuse_decode_rgba")
+                    self.generator.reconnect_input(pipeline_data, "layer_diffuse_decode_rgba.images", "vae_decode")
+                    if model_details.get("baseline") == "stable diffusion 1":
+                        pipeline_params["layer_diffuse_apply.config"] = "SD15, Attention Injection, attn_sharing"
+                        pipeline_params["layer_diffuse_decode_rgba.sd_version"] = "SD15"
+                    else:
+                        pipeline_params["layer_diffuse_apply.config"] = "SDXL, Conv Injection"
+                        pipeline_params["layer_diffuse_decode_rgba.sd_version"] = "SDXL"
+
         return pipeline_params, faults
 
     def _get_appropriate_pipeline(self, params):

hordelib/pipelines/pipeline_stable_diffusion.json

@@ -1,7 +1,7 @@
 {
   "3": {
     "inputs": {
-      "seed": 62706718437716,
+      "seed": 325847265780417,
       "steps": 20,
       "cfg": 8,
       "sampler_name": "euler",
@@ -155,5 +155,37 @@
     "_meta": {
       "title": "repeat_image_batch"
     }
+  },
+  "16": {
+    "inputs": {
+      "config": "SD15, Attention Injection, attn_sharing",
+      "weight": 1,
+      "model": [
+        "4",
+        0
+      ]
+    },
+    "class_type": "LayeredDiffusionApply",
+    "_meta": {
+      "title": "layer_diffuse_apply"
+    }
+  },
+  "17": {
+    "inputs": {
+      "sd_version": "SD15",
+      "sub_batch_size": 16,
+      "samples": [
+        "3",
+        0
+      ],
+      "images": [
+        "14",
+        0
+      ]
+    },
+    "class_type": "LayeredDiffusionDecodeRGBA",
+    "_meta": {
+      "title": "layer_diffuse_decode_rgba"
+    }
   }
 }

pyproject.toml

@@ -77,7 +77,7 @@ exclude = '''
 
 [tool.ruff] # XXX this isn't part of CI yet
 line-length=119
-exclude=["ComfyUI", "comfy_controlnet_preprocessors", "facerestore", "comfy_qr", "build"]
+exclude=["ComfyUI", "comfy_controlnet_preprocessors", "facerestore", "comfy_qr", "comfyui-layerdiffuse", "build"]
 ignore=[
     # "F401", # imported but unused
     "E402", # Module level import not at top of file

requirements.txt

@@ -3,6 +3,7 @@
 horde_sdk>=0.9.1
 horde_model_reference>=0.5.2
 pydantic
+numpy==1.26.4
 torch>=2.1.0
 # xformers>=0.0.19
 torchvision

tests/test_horde_inference_layerdiffusion.py

@@ -0,0 +1,92 @@
+# test_horde.py
+
+from PIL import Image
+
+from hordelib.horde import HordeLib
+
+
+class TestHordeInferenceTransparent:
+    def test_layerdiffuse_sd15(
+        self,
+        hordelib_instance: HordeLib,
+        stable_diffusion_model_name_for_testing: str,
+    ):
+        data = {
+            "sampler_name": "k_euler",
+            "cfg_scale": 8,
+            "denoising_strength": 1.0,
+            "seed": 1312,
+            "height": 512,
+            "width": 512,
+            "karras": False,
+            "tiling": False,
+            "hires_fix": False,
+            "transparent": True,
+            "clip_skip": 1,
+            "control_type": None,
+            "image_is_control": False,
+            "return_control_map": False,
+            "prompt": "an ancient digital AI hydra monster###watermark, text",
+            "ddim_steps": 25,
+            "n_iter": 2,
+            "model": stable_diffusion_model_name_for_testing,
+        }
+        image_results = hordelib_instance.basic_inference(data)
+
+        assert len(image_results) == 2
+
+        img_pairs_to_check = []
+
+        img_filename_base = "layer_diffusion_sd15_n_iter_{0}.png"
+
+        for i, image_result in enumerate(image_results):
+            assert image_result.image is not None
+            assert isinstance(image_result.image, Image.Image)
+
+            img_filename = img_filename_base.format(i)
+
+            image_result.image.save(f"images/{img_filename}", quality=100)
+            img_pairs_to_check.append((f"images_expected/{img_filename}", image_result.image))
+
+    def test_layerdiffuse_sdxl(
+        self,
+        hordelib_instance: HordeLib,
+        sdxl_refined_model_name: str,
+    ):
+        data = {
+            "sampler_name": "k_euler",
+            "cfg_scale": 8,
+            "denoising_strength": 1.0,
+            "seed": 1312,
+            "height": 1024,
+            "width": 1024,
+            "karras": False,
+            "tiling": False,
+            "hires_fix": False,
+            "transparent": True,
+            "clip_skip": 1,
+            "control_type": None,
+            "image_is_control": False,
+            "return_control_map": False,
+            "prompt": "an ancient digital AI hydra monster###watermark, text",
+            "ddim_steps": 25,
+            "n_iter": 2,
+            "model": sdxl_refined_model_name,
+        }
+
+        image_results = hordelib_instance.basic_inference(data)
+
+        assert len(image_results) == 2
+
+        img_pairs_to_check = []
+
+        img_filename_base = "layer_diffusion_sdxl_n_iter_{0}.png"
+
+        for i, image_result in enumerate(image_results):
+            assert image_result.image is not None
+            assert isinstance(image_result.image, Image.Image)
+
+            img_filename = img_filename_base.format(i)
+
+            image_result.image.save(f"images/{img_filename}", quality=100)
+            img_pairs_to_check.append((f"images_expected/{img_filename}", image_result.image))