control fix input size and image mode

CHANGELOG.md

@@ -3,10 +3,12 @@
 ## Update for 2023-12-30
 
 - **Fixes**:
-  - img2img clip and blip interrogate
-  - guard against invalid sampler index
-  - reset default cfg scale to 6.0
-  - processing tab display metadata
+  - control: fix input image size  
+  - control: fix correct image mode  
+  - img2img: clip and blip interrogate  
+  - guard against invalid sampler index  
+  - reset default cfg scale to 6.0  
+  - processing: correct display metadata  
 
 ## Update for 2023-12-29
 

modules/control/proc/canny.py

@@ -31,6 +31,5 @@ class CannyDetector:
 
         if output_type == "pil":
             detected_map = Image.fromarray(detected_map)
-            detected_map = detected_map.convert('L')
 
         return detected_map

modules/control/proc/edge.py

@@ -59,6 +59,5 @@ class EdgeDetector:
 
         if output_type == "pil":
             edge_map = Image.fromarray(edge_map)
-            edge_map = edge_map.convert('L')
 
         return edge_map

modules/control/processors.py

@@ -1,9 +1,9 @@
 import os
 import time
-import torch
 from PIL import Image
 from modules.shared import log
 from modules.errors import display
+from modules import devices
 
 from modules.control.proc.hed import HEDdetector
 from modules.control.proc.canny import CannyDetector
@@ -184,7 +184,7 @@ class Processor():
             display(e, 'Control processor load')
             return f'Processor load filed: {processor_id}'
 
-    def __call__(self, image_input: Image):
+    def __call__(self, image_input: Image, mode: str = 'RGB'):
         if self.override is not None:
             image_input = self.override
         image_process = image_input
@@ -203,19 +203,20 @@ class Processor():
             t0 = time.time()
             kwargs = config.get(self.processor_id, {}).get('params', None)
             if self.resize:
-                orig_size = image_input.size
                 image_resized = image_input.resize((512, 512))
             else:
                 image_resized = image_input
-            with torch.no_grad():
+            with devices.inference_context():
                 image_process = self.model(image_resized, **kwargs)
-            if self.resize:
-                image_process = image_process.resize(orig_size, Image.Resampling.LANCZOS)
+            if self.resize and image_process.size != image_input.size:
+                image_process = image_process.resize(image_input.size, Image.Resampling.LANCZOS)
             t1 = time.time()
-            log.debug(f'Control processor: id="{self.processor_id}" args={kwargs} time={t1-t0:.2f}')
+            log.debug(f'Control processor: id="{self.processor_id}" mode={mode} args={kwargs} time={t1-t0:.2f}')
         except Exception as e:
             log.error(f'Control processor failed: id="{self.processor_id}" error={e}')
             display(e, 'Control processor')
+        if mode != 'RGB':
+            image_process = image_process.convert(mode)
         return image_process
 
     def preview(self, image_input: Image):

modules/control/run.py

@@ -340,6 +340,9 @@ def control_run(units: List[unit.Unit], inputs, inits, unit_type: str, is_genera
                     if p.resize_mode != 0 and input_image is not None and resize_time == 'Before':
                         debug(f'Control resize: image={input_image} width={width} height={height} mode={p.resize_mode} name={resize_name} sequence={resize_time}')
                         input_image = images.resize_image(p.resize_mode, input_image, width, height, resize_name)
+                    p.width = input_image.width
+                    p.height = input_image.height
+                    debug(f'Control: input image={input_image}')
 
                     # process
                     if input_image is None:
@@ -357,7 +360,8 @@ def control_run(units: List[unit.Unit], inputs, inits, unit_type: str, is_genera
                             return msg
                         processed_image = p.ref_image
                     elif len(active_process) == 1:
-                        p.image = active_process[0](input_image)
+                        image_mode = 'L' if unit_type == 'adapter' and len(active_model) > 0 and ('Canny' in active_model[0].model_id or 'Sketch' in active_model[0].model_id) else 'RGB'
+                        p.image = active_process[0](input_image, image_mode)
                         p.task_args['image'] = p.image
                         p.extra_generation_params["Control process"] = active_process[0].processor_id
                         debug(f'Control: process={active_process[0].processor_id} image={p.image}')
@@ -369,7 +373,10 @@ def control_run(units: List[unit.Unit], inputs, inits, unit_type: str, is_genera
                         processed_image = p.image
                     else:
                         if len(active_process) > 0:
-                            p.image = [p(input_image) for p in active_process] # list[image]
+                            p.image = []
+                            for i, process in enumerate(active_process): # list[image]
+                                image_mode = 'L' if unit_type == 'adapter' and len(active_model) > i and ('Canny' in active_model[i].model_id or 'Sketch' in active_model[i].model_id) else 'RGB'
+                                p.image.append(process(input_image, image_mode))
                         else:
                             p.image = [input_image]
                         p.task_args['image'] = p.image

modules/processing_diffusers.py

@@ -139,6 +139,8 @@ def process_diffusers(p: StableDiffusionProcessing):
     def task_specific_kwargs(model):
         task_args = {}
         is_img2img_model = bool('Zero123' in shared.sd_model.__class__.__name__)
+        if len(getattr(p, 'init_images' ,[])) > 0:
+            p.init_images = [p.convert('RGB') for p in p.init_images]
         if sd_models.get_diffusers_task(model) == sd_models.DiffusersTaskType.TEXT_2_IMAGE and not is_img2img_model:
             p.ops.append('txt2img')
             if hasattr(p, 'width') and hasattr(p, 'height'):