remove-background: ben2

Signed-off-by: Vladimir Mandic <mandic00@live.com>

CHANGELOG.md

@@ -1,6 +1,6 @@
 # Change Log for SD.Next
 
-## Update for 2025-02-02
+## Update for 2025-02-03
 
 - **GitHub**
   - rename core repo from <https://github.com/vladmandic/automatic> to <https://github.com/vladmandic/sdnext>  
@@ -29,7 +29,10 @@
     with detailed defaults for each model type also configurable  
   - select between 150+ *OpenCLiP* supported models, 20+ built-in *VLMs*, *DeepDanbooru*  
   - **VLM**: now that we can use VLMs freely, we've also added support for few more out-of-the-box  
-    *Alibaba Qwen VL2*, *Huggingface Smol VL2*, *ToriiGate 0.4*  
+    [Alibaba Qwen VL2](https://huggingface.co/Qwen/Qwen2-VL-2B), [Huggingface Smol VL2](HuggingFaceTB/SmolVLM-Instruct), [ToriiGate 0.4](Minthy/ToriiGate-v0.4-2B)  
+- **Postprocess**  
+  - new sota remove background model: [BEN2](https://huggingface.co/PramaLLC/BEN2)  
+    select in *process -> remove background* or enable postprocessing for txt2img/img2img operations  
 - **Other**:
   - **networks**: imporove search/filter and add visual indicators for types  
   - **balanced offload** new defaults: *lowvram/4gb min threshold: 0, medvram/8gb min threshold: 0, default min threshold 0.25*  

modules/interrogate/vqa.py

@@ -9,8 +9,8 @@ from PIL import Image
 from modules import shared, devices, errors
 
 # TODO add additional vlmn
-# https://huggingface.co/nvidia/Eagle2-1B
-# https://huggingface.co/deepseek-ai/deepseek-vl2-tiny
+# https://huggingface.co/nvidia/Eagle2-1B not compatible with latest transformers
+# https://huggingface.co/deepseek-ai/deepseek-vl2-tiny requires custom code
 
 
 processor = None
@@ -35,7 +35,7 @@ vlm_models = {
     "Microsoft GIT TextCaps Base": "microsoft/git-base-textcaps", # 0.7GB
     "Microsoft GIT VQA Base": "microsoft/git-base-vqav2", # 0.7GB
     "Microsoft GIT VQA Large": "microsoft/git-large-vqav2", # 1.6GB
-    "ToriiGate 0.4 2B": "Minthy/ToriiGate-v0.4-2B", # TODO
+    "ToriiGate 0.4 2B": "Minthy/ToriiGate-v0.4-2B",
     "ViLT Base": "dandelin/vilt-b32-finetuned-vqa", # 0.5GB
 }
 vlm_prompts = [
@@ -122,10 +122,6 @@ def qwen(question: str, image: Image.Image, repo: str = None):
     return response
 
 
-def deepseek(question: str, image: Image.Image, repo: str = None):
-    return ''
-
-
 def smol(question: str, image: Image.Image, repo: str = None):
     global processor, model, loaded # pylint: disable=global-statement
     if model is None or loaded != repo:
@@ -347,8 +343,6 @@ def interrogate(question, image, model_name):
             answer = qwen(question, image, vqa_model)
         elif 'smol' in vqa_model.lower():
             answer = smol(question, image, vqa_model)
-        elif 'deepseek' in vqa_model.lower():
-            answer = deepseek(question, image, vqa_model)
         else:
             answer = 'unknown model'
     except Exception as e:

modules/scripts_postprocessing.py

@@ -4,9 +4,9 @@ from modules import errors, shared
 
 
 class PostprocessedImage:
-    def __init__(self, image):
+    def __init__(self, image, info = {}):
         self.image = image
-        self.info = {}
+        self.info = info
 
 
 class ScriptPostprocessing: