lollms-client 1.5.8__py3-none-any.whl → 1.5.9__py3-none-any.whl

lollms_client/__init__.py

@@ -8,7 +8,7 @@ from lollms_client.lollms_utilities import PromptReshaper # Keep general utiliti
 from lollms_client.lollms_mcp_binding import LollmsMCPBinding, LollmsMCPBindingManager
 from lollms_client.lollms_llm_binding import LollmsLLMBindingManager
 
-__version__ = "1.5.8" # Updated version
+__version__ = "1.5.9" # Updated version
 
 # Optionally, you could define __all__ if you want to be explicit about exports
 __all__ = [

lollms_client/tti_bindings/diffusers/__init__.py

@@ -35,7 +35,8 @@ try:
         AutoPipelineForInpainting,
         DiffusionPipeline,
         StableDiffusionPipeline,
-        
+        QwenImageEditPipeline,
+        QwenImageEditPlusPipeline
     )
     from diffusers.utils import load_image
     from PIL import Image
@@ -47,6 +48,8 @@ except ImportError:
     AutoPipelineForInpainting = None
     DiffusionPipeline = None
     StableDiffusionPipeline = None
+    QwenImageEditPipeline = None
+    QwenImageEditPlusPipeline = None
     Image = None
     load_image = None
     DIFFUSERS_AVAILABLE = False
@@ -359,7 +362,12 @@ class ModelManager:
                     common_args["safety_checker"] = None
                 if self.config.get("hf_cache_path"):
                     common_args["cache_dir"] = str(self.config["hf_cache_path"])
-                if task == "text2image":
+
+                if "Qwen-Image-Edit-2509" in str(model_path):
+                    self.pipeline = QwenImageEditPlusPipeline.from_pretrained(model_path, **common_args)
+                elif "Qwen-Image-Edit" in str(model_path):
+                    self.pipeline = QwenImageEditPipeline.from_pretrained(model_path, **common_args)
+                elif task == "text2image":
                     self.pipeline = AutoPipelineForText2Image.from_pretrained(model_path, **common_args)
                 elif task == "image2image":
                     self.pipeline = AutoPipelineForImage2Image.from_pretrained(model_path, **common_args)
@@ -498,9 +506,10 @@ class DiffusersTTIBinding_Impl(LollmsTTIBinding):
         {"family": "SD 1.x", "model_name": "runwayml/stable-diffusion-v1-5", "display_name": "Stable Diffusion 1.5", "desc": "Classic SD1.5."},
         {"family": "SD 2.x", "model_name": "stabilityai/stable-diffusion-2-1", "display_name": "Stable Diffusion 2.1", "desc": "SD2.1 base."},
         {"family": "SD3", "model_name": "stabilityai/stable-diffusion-3-medium-diffusers", "display_name": "Stable Diffusion 3 Medium", "desc": "SD3 medium."},
-        {"family": "Qwen", "model_name": "Qwen/Qwen-Image", "display_name": "Qwen Image Edit", "desc": "Dedicated image generation."},
+        {"family": "Qwen", "model_name": "Qwen/Qwen-Image", "display_name": "Qwen Image", "desc": "Dedicated image generation."},
         {"family": "Specialized", "model_name": "playgroundai/playground-v2.5-1024px-aesthetic", "display_name": "Playground v2.5", "desc": "High aesthetic 1024."},
-        {"family": "Editors", "model_name": "Qwen/Qwen-Image-Edit", "display_name": "Qwen Image Edit", "desc": "Dedicated image editing."}
+        {"family": "Editors", "model_name": "Qwen/Qwen-Image-Edit", "display_name": "Qwen Image Edit", "desc": "Dedicated image editing."},
+        {"family": "Editors", "model_name": "Qwen/Qwen-Image-Edit-2509", "display_name": "Qwen Image Edit Plus (Multi-Image)", "desc": "Advanced multi-image editing, fusion, and pose transfer."}
     ]
 
     def __init__(self, **kwargs):
@@ -679,6 +688,25 @@ class DiffusersTTIBinding_Impl(LollmsTTIBinding):
         generator = self._prepare_seed(kwargs)
         steps = kwargs.pop("num_inference_steps", self.config["num_inference_steps"])
         guidance = kwargs.pop("guidance_scale", self.config["guidance_scale"])
+
+        # Handle multi-image fusion for Qwen-Image-Edit-2509
+        if "Qwen-Image-Edit-2509" in self.model_name and len(pil_images) > 1:
+            pipeline_args = {
+                "image": pil_images,
+                "prompt": prompt,
+                "negative_prompt": negative_prompt or " ",
+                "width": out_w, "height": out_h,
+                "num_inference_steps": steps,
+                "true_cfg_scale": guidance,
+                "generator": generator
+            }
+            pipeline_args.update(kwargs)
+            future = Future()
+            self.manager.queue.put((future, "image2image", pipeline_args))
+            ASCIIColors.info(f"Job (multi-image fusion with {len(pil_images)} images) queued.")
+            return future.result()
+
+        # Handle inpainting (single image with mask)
         if mask is not None and len(pil_images) == 1:
             try:
                 mask_img = self._decode_image_input(mask).convert("L")
@@ -689,36 +717,46 @@ class DiffusersTTIBinding_Impl(LollmsTTIBinding):
                 "mask_image": mask_img,
                 "prompt": prompt,
                 "negative_prompt": negative_prompt or None,
-                "width": out_w,
-                "height": out_h,
+                "width": out_w, "height": out_h,
                 "num_inference_steps": steps,
                 "guidance_scale": guidance,
                 "generator": generator
             }
             pipeline_args.update(kwargs)
+            if "Qwen-Image-Edit" in self.model_name:
+                pipeline_args["true_cfg_scale"] = pipeline_args.pop("guidance_scale", 7.0)
+                if not pipeline_args.get("negative_prompt"): pipeline_args["negative_prompt"] = " "
+
             future = Future()
             self.manager.queue.put((future, "inpainting", pipeline_args))
             ASCIIColors.info("Job (inpaint) queued.")
             return future.result()
+
+        # Handle standard image-to-image (single image)
         try:
             pipeline_args = {
-                "image": pil_images if len(pil_images) > 1 else pil_images[0],
+                "image": pil_images[0],
                 "prompt": prompt,
                 "negative_prompt": negative_prompt or None,
                 "strength": kwargs.pop("strength", 0.6),
-                "width": out_w,
-                "height": out_h,
+                "width": out_w, "height": out_h,
                 "num_inference_steps": steps,
                 "guidance_scale": guidance,
                 "generator": generator
             }
             pipeline_args.update(kwargs)
+            if "Qwen-Image-Edit" in self.model_name:
+                pipeline_args["true_cfg_scale"] = pipeline_args.pop("guidance_scale", 7.0)
+                if not pipeline_args.get("negative_prompt"): pipeline_args["negative_prompt"] = " "
+
             future = Future()
             self.manager.queue.put((future, "image2image", pipeline_args))
             ASCIIColors.info("Job (i2i) queued.")
             return future.result()
         except Exception:
             pass
+
+        # Fallback to latent-based generation if i2i fails for some reason
         try:
             base = pil_images[0]
             latents, _ = self._encode_image_to_latents(base, out_w, out_h)
@@ -729,8 +767,7 @@ class DiffusersTTIBinding_Impl(LollmsTTIBinding):
                 "num_inference_steps": steps,
                 "guidance_scale": guidance,
                 "generator": generator,
-                "width": out_w,
-                "height": out_h
+                "width": out_w, "height": out_h
             }
             pipeline_args.update(kwargs)
             future = Future()
@@ -740,6 +777,7 @@ class DiffusersTTIBinding_Impl(LollmsTTIBinding):
         except Exception as e:
             raise Exception(f"Image edit failed: {e}") from e
 
+
     def list_local_models(self) -> List[str]:
         if not self.models_path.exists():
             return []
@@ -846,4 +884,4 @@ if __name__ == '__main__':
         ASCIIColors.cyan("\nCleaning up temporary directories...")
         if temp_paths_dir.exists():
             shutil.rmtree(temp_paths_dir)
-        ASCIIColors.magenta("--- Diffusers TTI Binding Test Finished ---")
+        ASCIIColors.magenta("--- Diffusers TTI Binding Test Finished ---")

{lollms_client-1.5.8.dist-info → lollms_client-1.5.9.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: lollms_client
-Version: 1.5.8
+Version: 1.5.9
 Summary: A client library for LoLLMs generate endpoint
 Author-email: ParisNeo <parisneoai@gmail.com>
 License:                                  Apache License
@@ -249,7 +249,8 @@ Whether you're connecting to a remote LoLLMs server, an Ollama instance, the Ope
 ## Key Features
 
 *   🔌 **Versatile Binding System:** Seamlessly switch between different LLM backends (LoLLMs, Ollama, OpenAI, Llama.cpp, Transformers, vLLM, OpenLLM, Gemini, Claude, Groq, OpenRouter, Hugging Face Inference API) using a unified `llm_binding_config` dictionary for all parameters.
-*   🗣️ **Multimodal Support:** Interact with models capable of processing images and generate various outputs like speech (TTS), images (TTI), video (TTV), and music (TTM).
+*   🗣️ **Comprehensive Multimodal Support:** Interact with models capable of processing images and generate various outputs like speech (TTS), video (TTV), and music (TTM).
+*   🎨 **Advanced Image Generation and Editing:** A new `diffusers` binding provides powerful text-to-image capabilities. It supports a wide range of models from Hugging Face and Civitai, including specialized models like `Qwen-Image-Edit` for single-image editing and the cutting-edge `Qwen-Image-Edit-2509` for **multi-image fusion, pose transfer, and character swapping**.
 *   🖼️ **Selective Image Activation:** Control which images in a message are active and sent to the model, allowing for fine-grained multimodal context management without deleting the original data.
 *   🤖 **Agentic Workflows with MCP:** Empower LLMs to act as sophisticated agents, breaking down complex tasks, selecting and executing external tools (e.g., internet search, code interpreter, file I/O, image generation) through the Model Context Protocol (MCP) using a robust "observe-think-act" loop.
 *   🎭 **Personalities as Agents:** Personalities can now define their own set of required tools (MCPs) and have access to static or dynamic knowledge bases (`data_source`), turning them into self-contained, ready-to-use agents.
@@ -1300,9 +1301,110 @@ try:
 
 except Exception as e:
     ASCIIColors.error(f"Error initializing Hugging Face Inference API binding: {e}")
-    ASCIIColors.info("Please ensure your Hugging Face API token is correctly set and you have access to the specified model.")
+    ASCIIColors.info("Please ensure your Hugging Face API token is correctly set and you have access to the specified model.")```
+
+---
+
+### 4. Local Multimodal and Advanced Bindings
+
+#### **Diffusers (Local Text-to-Image Generation and Editing)**
+
+The `diffusers` binding leverages the Hugging Face `diffusers` library to run a vast array of text-to-image models locally on your own hardware (CPU or GPU). It supports models from Hugging Face and Civitai, providing everything from basic image generation to advanced, state-of-the-art image editing.
+
+**Prerequisites:**
+*   `torch` and `torchvision` must be installed. For GPU acceleration, it's critical to install the version that matches your CUDA toolkit.
+*   The binding will attempt to auto-install other requirements like `diffusers`, `transformers`, and `safetensors`.
+
+**Usage:**
+
+**Example 1: Basic Text-to-Image Generation**
+This example shows how to generate an image from a simple text prompt using a classic Stable Diffusion model.
+
+```python
+from lollms_client import LollmsClient
+from ascii_colors import ASCIIColors
+from pathlib import Path
+
+try:
+    # Initialize the client with the diffusers TTI binding
+    # Let's use a classic Stable Diffusion model for this example
+    lc = LollmsClient(
+        tti_binding_name="diffusers",
+        tti_binding_config={
+            "model_name": "runwayml/stable-diffusion-v1-5",
+            # Other options: "device", "torch_dtype_str", "enable_xformers"
+        }
+    )
+
+    prompt = "A high-quality photograph of an astronaut riding a horse on Mars."
+    ASCIIColors.yellow(f"Generating image for prompt: '{prompt}'")
+
+    # Generate the image. The result is returned as bytes.
+    image_bytes = lc.generate_image(prompt, width=512, height=512)
+
+    if image_bytes:
+        output_path = Path("./astronaut_on_mars.png")
+        with open(output_path, "wb") as f:
+            f.write(image_bytes)
+        ASCIIColors.green(f"Image saved successfully to: {output_path.resolve()}")
+    else:
+        ASCIIColors.error("Image generation failed.")
+
+except Exception as e:
+    ASCIIColors.error(f"An error occurred with the Diffusers binding: {e}")
+    ASCIIColors.info("Please ensure torch is installed correctly for your hardware (CPU/GPU).")
 ```
 
+**Example 2: Advanced Multi-Image Fusion with Qwen-Image-Edit-2509**
+This example demonstrates a cutting-edge capability: using a specialized model to fuse elements from multiple input images based on a text prompt. Here, we'll ask the model to take a person from one image and place them in the background of another.
+
+```python
+from lollms_client import LollmsClient
+from ascii_colors import ASCIIColors
+from pathlib import Path
+
+# --- IMPORTANT ---
+# Replace these with actual paths to your local images
+path_to_person_image = "./path/to/your/person.jpg"
+path_to_background_image = "./path/to/your/background.jpg"
+
+if not Path(path_to_person_image).exists() or not Path(path_to_background_image).exists():
+    ASCIIColors.warning("Input images not found. Skipping multi-image fusion example.")
+    ASCIIColors.warning(f"Please update 'path_to_person_image' and 'path_to_background_image'.")
+else:
+    try:
+        # Initialize with the advanced Qwen multi-image editing model
+        lc = LollmsClient(
+            tti_binding_name="diffusers",
+            tti_binding_config={
+                "model_name": "Qwen/Qwen-Image-Edit-2509",
+                "torch_dtype_str": "bfloat16" # Recommended for this model
+            }
+        )
+
+        # The prompt guides how the images are combined
+        prompt = "Place the person from the first image into the scenic background of the second image."
+        ASCIIColors.yellow(f"Fusing images with prompt: '{prompt}'")
+
+        # The edit_image method can accept a list of image paths for fusion
+        fused_image_bytes = lc.edit_image(
+            images=[path_to_person_image, path_to_background_image],
+            prompt=prompt,
+            num_inference_steps=50
+        )
+
+        if fused_image_bytes:
+            output_path = Path("./fused_image_result.png")
+            with open(output_path, "wb") as f:
+                f.write(fused_image_bytes)
+            ASCIIColors.green(f"Fused image saved successfully to: {output_path.resolve()}")
+        else:
+            ASCIIColors.error("Multi-image editing failed.")
+
+    except Exception as e:
+        ASCIIColors.error(f"An error occurred during multi-image fusion: {e}")
+```This powerful feature allows for complex creative tasks like character swapping, background replacement, and style transfer directly through the `lollms_client` library.
+
 ### Listing Available Models
 
 You can query the active LLM binding to get a list of models it supports or has available. The exact information returned depends on the binding (e.g., Ollama lists local models, OpenAI lists all its API models).

{lollms_client-1.5.8.dist-info → lollms_client-1.5.9.dist-info}/RECORD

@@ -1,4 +1,4 @@
-lollms_client/__init__.py,sha256=LmAWm88VN_dcmwMvHFYF3nl_p7TqW2D6oI1xjWmhXJ0,1146
+lollms_client/__init__.py,sha256=pBf8IT3SoW2Y-kWsYgmN-bq3gC70uJf6m9ZjNqzg4IE,1146
 lollms_client/lollms_agentic.py,sha256=pQiMEuB_XkG29-SW6u4KTaMFPr6eKqacInggcCuCW3k,13914
 lollms_client/lollms_config.py,sha256=goEseDwDxYJf3WkYJ4IrLXwg3Tfw73CXV2Avg45M_hE,21876
 lollms_client/lollms_core.py,sha256=Gc2KMyxnKWti-ljI7rf3XDb6a7NM2RUpktN8VnlWcyk,323791
@@ -52,7 +52,7 @@ lollms_client/stt_bindings/lollms/__init__.py,sha256=9Vmn1sQQZKLGLe7nZnc-0LnNeSY
 lollms_client/stt_bindings/whisper/__init__.py,sha256=1Ej67GdRKBy1bba14jMaYDYHiZkxJASkWm5eF07ztDQ,15363
 lollms_client/stt_bindings/whispercpp/__init__.py,sha256=xSAQRjAhljak3vWCpkP0Vmdb6WmwTzPjXyaIB85KLGU,21439
 lollms_client/tti_bindings/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-lollms_client/tti_bindings/diffusers/__init__.py,sha256=3ANfQ3WtOpyMmxdfQDX4mteupZ6sLFlE-GVj-Qi-FVI,41849
+lollms_client/tti_bindings/diffusers/__init__.py,sha256=IB4yZpYOY7bq9l1rRNM9SZzy5EbKqGsY2BlEY3hX_Pg,43892
 lollms_client/tti_bindings/gemini/__init__.py,sha256=f9fPuqnrBZ1Z-obcoP6EVvbEXNbNCSg21cd5efLCk8U,16707
 lollms_client/tti_bindings/leonardo_ai/__init__.py,sha256=EvjKyV8kM7-tmLwC1agSQ-v7thrNgflrmFxhDLqzT8U,5884
 lollms_client/tti_bindings/lollms/__init__.py,sha256=5Tnsn4b17djvieQkcjtIDBm3qf0pg5ZWWov-4_2wmo0,8762
@@ -80,8 +80,8 @@ lollms_client/tts_bindings/xtts/server/main.py,sha256=T-Kn5NM-u1FJMygeV8rOoZKlqn
 lollms_client/tts_bindings/xtts/server/setup_voices.py,sha256=UdHaPa5aNcw8dR-aRGkZr2OfSFFejH79lXgfwT0P3ss,1964
 lollms_client/ttv_bindings/__init__.py,sha256=UZ8o2izQOJLQgtZ1D1cXoNST7rzqW22rL2Vufc7ddRc,3141
 lollms_client/ttv_bindings/lollms/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-lollms_client-1.5.8.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
-lollms_client-1.5.8.dist-info/METADATA,sha256=EMkJabEuQ0JcBVg7KkQQzfTvrWO65N4-44qM8uEW4kQ,71861
-lollms_client-1.5.8.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-lollms_client-1.5.8.dist-info/top_level.txt,sha256=Bk_kz-ri6Arwsk7YG-T5VsRorV66uVhcHGvb_g2WqgE,14
-lollms_client-1.5.8.dist-info/RECORD,,
+lollms_client-1.5.9.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
+lollms_client-1.5.9.dist-info/METADATA,sha256=LZ3XLpnSfsEKZHgVa3HpIX3Didujw4Eo5AMIIwi_khc,76825
+lollms_client-1.5.9.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+lollms_client-1.5.9.dist-info/top_level.txt,sha256=Bk_kz-ri6Arwsk7YG-T5VsRorV66uVhcHGvb_g2WqgE,14
+lollms_client-1.5.9.dist-info/RECORD,,