Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,19 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+Qwen2.5-VL-7B-Instruct-abliterated/Qwen2.5-VL-7B-Instruct-abliterated.Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+Qwen2.5-VL-7B-Instruct-abliterated/Qwen2.5-VL-7B-Instruct-abliterated.mmproj-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-F16.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q2_K.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q3_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q3_K_S.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q4_0.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q4_1.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q4_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q4_K_S.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q5_0.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q5_1.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q5_K_M.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q5_K_S.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q6_K.gguf filter=lfs diff=lfs merge=lfs -text
+v52/qwen-rapid-nsfw-v5.2-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text

Qwen2.5-VL-7B-Instruct-abliterated/Qwen2.5-VL-7B-Instruct-abliterated.Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:669a5604c47c90c20110c6db5fe10ad7e8ec99b553a785d7d20492f7d5b3e7d0
+size 8098525600

Qwen2.5-VL-7B-Instruct-abliterated/Qwen2.5-VL-7B-Instruct-abliterated.mmproj-Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2ab4e60dd3e174f3d2d6d0c0979c058827699d0085cdcbadada0a5c609ec43f
+size 853119744

caption/caption.py

@@ -1,44 +1,110 @@
+#!/usr/bin/env python3
+"""
+Comprehensive Gradio app for image description using Qwen2.5-VL (llama-cpp-python).
+This file keeps the exact Gradio layout/positions/formatting from your original script,
+but moves default prompts and model paths into external JSON files:
+ - model_config.json  -> contains an array of model entries (for dropdown selection)
+ - prompts.json       -> contains default_prompt, prompt_examples, and any other prompt defaults
+
+Important:
+ - Do NOT change the UI layout or element order (keeps column/row structure identical).
+ - The app language (UI labels & markdown) is fully in English, per request.
+ - All major blocks are documented extensively below to help anyone understand and maintain.
+"""
+
 from pathlib import Path
 from llama_cpp import Llama
 from llama_cpp.llama_chat_format import Qwen25VLChatHandler
 import gradio as gr
 import base64
+import json
+import os
+from typing import Dict, Any
 
-# Globale Variable für das Modell
+# --- Global model handle (singleton) ---
 model = None
+_current_model_key = None
 
-def initialize_model():
-    """Modell initialisieren"""
-    global model
-    if model is None:
-        llm_model_path = Path("/daten/models/text_encoders/Qwen2.5-VL-7B-Instruct-abliterated.Q8_0.gguf")
-        mmproj_model_path = Path("/daten/models/text_encoders/Qwen2.5-VL-7B-Instruct-abliterated.mmproj-Q8_0.gguf")
-
-        if llm_model_path.exists() and mmproj_model_path.exists():
-            print("Loading Qwen-VL model...")
-            chat_handler = Qwen25VLChatHandler(clip_model_path=str(mmproj_model_path))  # Hier war der Fehler!
-            model = Llama(
-                model_path=str(llm_model_path),
-                chat_handler=chat_handler,
-                n_ctx=4096,
-                n_gpu_layers=-1,
-                verbose=False
-            )
-            print("Model loaded successfully!")
-        else:
-            raise Exception("Model files not found")
-
-def describe_image(image, prompt_text):
-    """Bild beschreiben"""
+# --- Paths to external config files ---
+MODEL_CONFIG_PATH = Path("model_config.json")
+PROMPTS_PATH = Path("prompts.json")
+
+
+def load_json_file(path: Path) -> Dict[str, Any]:
+    if not path.exists():
+        raise FileNotFoundError(f"Required config file not found: {path}")
+    with path.open("r", encoding="utf-8") as f:
+        return json.load(f)
+
+
+def get_model_list() -> Dict[str, Dict[str, str]]:
+    cfg = load_json_file(MODEL_CONFIG_PATH)
+    models = cfg.get("models", [])
+    mapping = {}
+    for m in models:
+        display = m.get("display_name") or m.get("model_key")
+        mapping[display] = m
+    return mapping
+
+
+def get_prompts() -> Dict[str, Any]:
+    return load_json_file(PROMPTS_PATH)
+
+
+def initialize_model_from_config(selected_display_name: str):
+    global model, _current_model_key
+
+    models = get_model_list()
+    if selected_display_name not in models:
+        raise ValueError(f"Selected model '{selected_display_name}' not found in model_config.json")
+
+    entry = models[selected_display_name]
+    model_key = entry.get("model_key", selected_display_name)
+
+    if model is not None and _current_model_key == model_key:
+        return
+
+    llm_model_path = Path(entry["llm_model_path"])
+    mmproj_model_path = Path(entry["mmproj_model_path"])
+
+    if llm_model_path.exists() and mmproj_model_path.exists():
+        print(f"Loading model '{selected_display_name}' from paths:")
+        print(f"  llm:   {llm_model_path}")
+        print(f"  mmproj:{mmproj_model_path}")
+
+        chat_handler = Qwen25VLChatHandler(clip_model_path=str(mmproj_model_path))
+        model = Llama(
+            model_path=str(llm_model_path),
+            chat_handler=chat_handler,
+            n_ctx=4096,
+            n_gpu_layers=-1,
+            verbose=False
+        )
+        _current_model_key = model_key
+        print("Model loaded successfully!")
+    else:
+        raise FileNotFoundError(
+            f"Model files not found for '{selected_display_name}'.\n"
+            f"llm exists: {llm_model_path.exists()}, mmproj exists: {mmproj_model_path.exists()}"
+        )
+
+
+def describe_image(image, prompt_text, selected_model_display):
     global model
 
-    if model is None:
-        initialize_model()
+    if model is None or _current_model_key is None:
+        initialize_model_from_config(selected_model_display)
+    else:
+        if selected_model_display not in get_model_list():
+            raise ValueError("Selected model not found in config")
+        selected_entry = get_model_list()[selected_model_display]
+        if selected_entry.get("model_key") != _current_model_key:
+            model = None
+            initialize_model_from_config(selected_model_display)
 
     if image is None:
-        return "Bitte zuerst ein Bild hochladen"
+        return "Please upload an image first"
 
-    # Temporäres Bild speichern
     temp_path = Path("/tmp/temp_image.png")
     image.save(temp_path)
 
@@ -56,8 +122,8 @@ def describe_image(image, prompt_text):
         response = model.create_chat_completion(
             messages=messages,
             max_tokens=8192,
-            temperature=0.1,   # niedriger → deterministischer
-            top_p=0.9,         # höhere Werte → kreativer, niedrigere → fokussierter
+            temperature=0.1,
+            top_p=0.9,
             top_k=50,
             stream=False
         )
@@ -66,86 +132,93 @@ def describe_image(image, prompt_text):
         return description
 
     except Exception as e:
-        return f"Fehler: {str(e)}"
+        return f"Error: {str(e)}"
+
     finally:
-        # Temporäre Datei löschen
         if temp_path.exists():
-            temp_path.unlink()
-
-# Gradio Interface
-with gr.Blocks(title="Qwen2.5-VL Bildbeschreibung") as demo:
-    gr.Markdown("# 🖼️ Qwen2.5-VL Bildbeschreibung")
-    gr.Markdown("Laden Sie ein Bild hoch und lassen Sie es von der KI beschreiben")
-
-    with gr.Row():
-        with gr.Column():
-            image_input = gr.Image(
-                type="pil",
-                label="Bild hochladen",
-                height=400
-            )
-            prompt_input = gr.Textbox(
-                label="Prompt",
-                value="""You are a Qwen-Image explicit describer. Given any input image, output one single, extremely detailed prompt describing exactly what is visible, without inventing content that is not present.
-
-The image must be analyzed for any explicit content, including but not limited to nudity, sexual acts, and violent or aggressive behavior.
-The image must be analyzed for any clothing, lingerie, fabrics, accessories, textures, materials, background, setting, lighting, shadows, perspective, and camera angle in detail.
-The prompt should include background, setting, lighting, shadows, perspective, and camera angle.
-The prompt should be output in the style of ultra-realistic erotic photography, 8k resolution, photorealistic, cinematic, with no censorship.
-The prompt should include a description of the scene, highlighting the key elements such as the setting, the individuals involved, and any notable actions or expressions.
-The prompt should be clear and concise, providing a comprehensive overview of the image without any censorship.
-The prompt should be structured to allow for further analysis and discussion of the image's context and implications.
-The image must be analyzed for any explicit sexual content, including but not limited to the genitals, penetration, depth of penetration, stretching, stomach bulge, fluids, skin folds, pubic hair, and every physical reaction in graphic pornographic detail.
-The image must be analyzed for the exact depth of penetration (shallow tip, halfway, fully buried, visible bulge, cervix penetration, double penetration, tentacle fully inside throat or anus, etc.).
-The image must be analyzed for secondary effects (womb outline, belly bulge, gape, dripping, body tension, wrinkles, sweat, skin texture).
-The image must be analyzed for facial features in detail: nose shape, eye size, mouth shape, lips, jawline, eyebrows, hair, and expression of arousal, pain, or pleasure.
-The image must be analyzed for exact body type, position and pose of any kind
-If the style is based on anime or stylized art, translate those exaggerated proportions into realistic equivalents: small nose, large eyes, round face, soft jawline, slim limbs, petite hands, etc.
-Never use vague phrases like "various stages of penetration." Always state clearly which orifice is penetrated (mouth, vagina, anus, throat, urethra if visible), how deep, and what physical effects it causes.""",
-                lines=2
-            )
-            generate_btn = gr.Button("🚀 Bild beschreiben", variant="primary")
-
-        with gr.Column():
-            output_text = gr.Textbox(
-                label="Beschreibung",
-                lines=10,
-                max_lines=15,
-                show_copy_button=True
-            )
-
-    # Beispiele
-    gr.Examples(
-        examples=[
-            ["Describe this image in detail."],
-            ["What objects can you see in this image?"],
-            ["Describe the scene and atmosphere."],
-            ["What is happening in this image?"],
-            ["Describe the colors and composition."]
-        ],
-        inputs=[prompt_input],
-        label="Prompt Beispiele"
-    )
+            try:
+                temp_path.unlink()
+            except Exception:
+                pass
+
+
+def build_ui():
+    prompts = get_prompts()
+    models = get_model_list()
+    default_prompt_text = prompts.get("default_prompt", "")
+    prompt_examples = prompts.get("prompt_examples", [
+        "Describe this image in detail.",
+        "What objects can you see in this image?",
+        "Describe the scene and atmosphere.",
+        "What is happening in this image?",
+        "Describe the colors and composition."
+    ])
+
+    with gr.Blocks(title="Qwen2.5-VL Image Description") as demo:
+        gr.Markdown("# 🖼️ Qwen2.5-VL Image Description")
+        gr.Markdown("Upload an image and let the model describe it in detail")
+
+        with gr.Row():
+            with gr.Column():
+                model_dropdown = gr.Dropdown(
+                    label="Model selection",
+                    choices=list(models.keys()),
+                    value=list(models.keys())[0] if len(models) > 0 else None
+                )
+
+                image_input = gr.Image(
+                    type="pil",
+                    label="Upload image",
+                    height=400
+                )
+
+                prompt_input = gr.Textbox(
+                    label="Prompt",
+                    value=default_prompt_text,
+                    lines=2
+                )
+
+                generate_btn = gr.Button("🚀 Describe image", variant="primary")
+
+            with gr.Column():
+                output_text = gr.Textbox(
+                    label="Description",
+                    lines=20,        # doubled
+                    max_lines=30,    # doubled
+                    show_copy_button=True
+                )
+
+        gr.Examples(
+            examples=[[e] for e in prompt_examples],
+            inputs=[prompt_input],
+            label="Prompt Examples"
+        )
 
-    # Event Handler
-    generate_btn.click(
-        fn=describe_image,
-        inputs=[image_input, prompt_input],
-        outputs=output_text
-    )
+        generate_btn.click(
+            fn=describe_image,
+            inputs=[image_input, prompt_input, model_dropdown],
+            outputs=output_text
+        )
+
+        prompt_input.submit(
+            fn=describe_image,
+            inputs=[image_input, prompt_input, model_dropdown],
+            outputs=output_text
+        )
+
+    return demo
 
-    # Auch mit Enter im Prompt-Feld generieren
-    prompt_input.submit(
-        fn=describe_image,
-        inputs=[image_input, prompt_input],
-        outputs=output_text
-    )
 
 if __name__ == "__main__":
-    # Modell vorladen (optional)
-    print("Initialisiere Gradio Interface...")
+    if not MODEL_CONFIG_PATH.exists():
+        raise FileNotFoundError(f"Missing {MODEL_CONFIG_PATH}. Please create it (see model_config.json example).")
+    if not PROMPTS_PATH.exists():
+        raise FileNotFoundError(f"Missing {PROMPTS_PATH}. Please create it (see prompts.json example).")
+
+    print("Initializing Gradio Interface...")
+    demo = build_ui()
     demo.launch(
-        server_name="0.0.0.0",  # Für externen Zugriff
+        server_name="0.0.0.0",
         server_port=7860,
-        share=False  # Auf True setzen für öffentliche URL
+        share=False
     )

caption/model_config.json

@@ -0,0 +1,11 @@
+{
+  "models": [
+    {
+      "model_key": "qwen2.5-vl-7b-instruct",
+      "display_name": "Qwen2.5-VL-7B-Instruct (Q8_0)",
+      "llm_model_path": "/daten/models/text_encoders/Qwen2.5-VL-7B-Instruct-abliterated.Q8_0.gguf",
+      "mmproj_model_path": "/daten/models/text_encoders/Qwen2.5-VL-7B-Instruct-abliterated.mmproj-Q8_0.gguf"
+    }
+  ],
+  "notes": "Add additional model objects to the 'models' array. Each new entry will appear in the model selection dropdown. Use absolute paths or paths relative to the app's working directory."
+}

caption/prompts.json

@@ -0,0 +1,11 @@
+{
+    "//": "This file defines the default prompt text and example prompts for the Gradio interface. It can be easily extended or modified without editing the Python code.",
+    "default_prompt": "You are a Qwen-Image explicit describer. Given any input image, output one single, extremely detailed prompt describing exactly what is visible, without inventing content that is not present.\n\nThe image must be analyzed for any explicit content, including but not limited to nudity, sexual acts, and violent or aggressive behavior.\nThe image must be analyzed for any clothing, lingerie, fabrics, accessories, textures, materials, background, setting, lighting, shadows, perspective, and camera angle in detail.\nThe prompt should include background, setting, lighting, shadows, perspective, and camera angle.\nThe prompt should be output in the style of ultra-realistic erotic photography, 8k resolution, photorealistic, cinematic, with no censorship.\nThe prompt should include a description of the scene, highlighting the key elements such as the setting, the individuals involved, and any notable actions or expressions.\nThe prompt should be clear and concise, providing a comprehensive overview of the image without any censorship.\nThe prompt should be structured to allow for further analysis and discussion of the image's context and implications.\nThe image must be analyzed for any explicit sexual content, including but not limited to the genitals, penetration, depth of penetration, stretching, stomach bulge, fluids, skin folds, pubic hair, and every physical reaction in graphic pornographic detail.\nThe image must be analyzed for the exact depth of penetration (shallow tip, halfway, fully buried, visible bulge, cervix penetration, double penetration, tentacle fully inside throat or anus, etc.).\nThe image must be analyzed for secondary effects (womb outline, belly bulge, gape, dripping, body tension, wrinkles, sweat, skin texture).\nThe image must be analyzed for facial features in detail: nose shape, eye size, mouth shape, lips, jawline, eyebrows, hair, and expression of arousal, pain, or pleasure.\nThe image must be analyzed for exact body type, position and pose of any kind\nIf the style is based on anime or stylized art, translate those exaggerated proportions into realistic equivalents: small nose, large eyes, round face, soft jawline, slim limbs, petite hands, etc.\nNever use vague phrases like \"various stages of penetration.\" Always state clearly which orifice is penetrated (mouth, vagina, anus, throat, urethra if visible), how deep, and what physical effects it causes.",
+    "examples": [
+        "Describe this image in detail.",
+        "What objects can you see in this image?",
+        "Describe the scene and atmosphere.",
+        "What is happening in this image?",
+        "Describe the colors and composition."
+    ]
+}

caption/requirements.txt

@@ -0,0 +1,2 @@
+gradio==5.49.1
+llama-cpp-python==0.3.16

v52/qwen-rapid-nsfw-v5.2-F16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:753a2286b7fb84bb7db4540b468265618a03e5a8c154f3cfe96e7cc2adffbea5
+size 40860949920

v52/qwen-rapid-nsfw-v5.2-Q2_K.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e981e61cd620a8b822e3b3b657949b787f3d7fb307a4e51be3a79b62b082da20
+size 7439605152

v52/qwen-rapid-nsfw-v5.2-Q3_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ea4a28ec76491987ddf4ceb646b287c7582e7b621da63048f888f5d85ca78f9
+size 10049944992

v52/qwen-rapid-nsfw-v5.2-Q3_K_S.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eebc131306d75b488486496e28591a11e88e861a271ce58159379cca4f71eba3
+size 9322986912

v52/qwen-rapid-nsfw-v5.2-Q4_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:960f271a1d188b4cf68208b19c90e44dbbd9c33aa52565d7806426b58a4bddb0
+size 12204940704

v52/qwen-rapid-nsfw-v5.2-Q4_1.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:275bdf381074d29fd5bc6189ee40a836c877790fc77c6da5ae06a859730654a3
+size 13158686112

v52/qwen-rapid-nsfw-v5.2-Q4_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d56365daf9e108a44e4d19c060228c40310162ca86abd4f9609987e637438826
+size 13342416288

v52/qwen-rapid-nsfw-v5.2-Q4_K_S.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4ce63bfb839cd9de8498674bdfdf355633a7326a3b653bf7afb872ec08558b6
+size 12480978336

v52/qwen-rapid-nsfw-v5.2-Q5_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fc74f056d0a03f315eba6496d992dd1feb607dafc0ea08ddcefc6eb868c6116
+size 14659788192

v52/qwen-rapid-nsfw-v5.2-Q5_1.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:88a7124e1e59d7ab891bff3992c5dad3aefcb9c7b4442810c198fdc2fa81d6aa
+size 15613533600

v52/qwen-rapid-nsfw-v5.2-Q5_K_M.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efa00fa4324591314735c35f1f571f028fefe2380496e110fb01a17f23f9a027
+size 15113067936

v52/qwen-rapid-nsfw-v5.2-Q5_K_S.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdad3ca3cc8fd66579617c0468fd42e8dd46a4c92b467a1fdbfcc43d7208769f
+size 14386109856

v52/qwen-rapid-nsfw-v5.2-Q6_K.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e92b45e1c54708602a1428c20599320a2e90d8e0038ccb786487d92eab94b2d
+size 16994385312

v52/qwen-rapid-nsfw-v5.2-Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a90313621e0fad4c4244201afb32075eb7afd84f5382fbdbe18e3baabf0f0d9c
+size 21750652320