Upload 2 files

README.md

@@ -23,11 +23,11 @@ You can call this Space from Python (via `gradio_client`) or from plain `curl`.
 > ⚠️ Note: This README may lag behind the actual API definition shown in the Space’s “View API” page.
 > If something does not work, always double-check the latest argument list and endpoint names there.
 
-All examples assume:
+Assumptions:
 
 - Space ID: `John6666/DiffuseCraftMod`
 - You have a valid Hugging Face access token: `hf_xxx...` (read access is enough)
-- You replace `hf_xxx...` with your own token
+- Replace `hf_xxx...` with your own token
 
 ---
 
@@ -47,27 +47,34 @@ from gradio_client import Client
 client = Client("John6666/DiffuseCraftMod", hf_token="hf_xxx...")
 
 status, images, info = client.predict(
+    # Core text controls
     prompt="Hello!!",
     negative_prompt=(
         "lowres, bad anatomy, bad hands, missing fingers, extra digit, "
         "fewer digits, worst quality, low quality"
     ),
+
+    # Basic generation controls
     num_images=1,
     num_inference_steps=28,
     guidance_scale=7.0,
     clip_skip=0,
     seed=-1,
+
+    # Canvas / model / task (optional, server has defaults)
     height=1024,
     width=1024,
     model_name="votepurchase/animagine-xl-3.1",
     vae_model="None",
     task="txt2img",
+
+    # All other arguments are optional; defaults match the UI
     api_name="/generate_image",
 )
 
-print(status)
-print(images)
-print(info)
+print(status)   # e.g. "COMPLETE"
+print(images)   # list of image paths / URLs
+print(info)     # generation metadata (seed, model, etc.)
 ```
 
 #### 1.2 Streaming API – `generate_image_stream`
@@ -101,6 +108,8 @@ for status, images, info in job:
     print(status, images, info)
 ```
 
+You can stop iterating once you see a `"COMPLETE"` status if you only care about the final output.
+
 ---
 
 ### 2. `curl` examples
@@ -111,6 +120,9 @@ When calling from `curl`, include your HF token; anonymous calls may be rate-lim
 export HF_TOKEN="hf_xxx..."  # your Hugging Face access token
 ```
 
+The `data` field is a positional array. The order must match the function signature.
+For simplicity, the examples below only send the first few arguments and rely on server defaults for the rest.
+
 #### 2.1 Synchronous API – `generate_image`
 
 ```bash
@@ -119,18 +131,14 @@ curl -X POST "https://john6666-diffusecraftmod.hf.space/call/generate_image" \
   -H "Content-Type: application/json" \
   -d '{
     "data": [
-      "Hello!!",
-      "lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, worst quality, low quality",
-      1,
-      28,
-      7.0,
-      0,
-      -1,
-      1024,
-      1024,
-      "votepurchase/animagine-xl-3.1",
-      "None",
-      "txt2img"
+      "Hello!!",  // prompt
+      "lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, worst quality, low quality", // negative_prompt
+      1,          // num_images
+      28,         // num_inference_steps
+      7.0,        // guidance_scale
+      0,          // clip_skip
+      -1          // seed
+      // All subsequent parameters will use their default values
     ]
   }'
 ```
@@ -149,15 +157,10 @@ curl -X POST "https://john6666-diffusecraftmod.hf.space/call/generate_image_stre
       28,
       7.0,
       0,
-      -1,
-      1024,
-      1024,
-      "votepurchase/animagine-xl-3.1",
-      "None",
-      "txt2img"
+      -1
     ]
   }'
 ```
 
-For full parameter coverage (all advanced options), refer to the Space’s “View API” page and adapt the
-examples above accordingly.
+For full parameter coverage (all advanced options such as LoRAs, ControlNet, IP-Adapter, etc.),
+refer to the Space’s “View API” page and adapt the examples above accordingly.

app.py

@@ -1887,9 +1887,19 @@ with gr.Blocks(theme=args.theme, elem_id="main", fill_width=True, fill_height=Fa
         copy_prompt_btn_pony.click(gradio_copy_prompt, inputs=[output_text_pony], outputs=[prompt_gui], show_api=False)
 
         from typing import Any, Dict, List, Optional, Tuple, Generator
-        import inspect
-        # 1) Single source of truth: define the full explicit signature once.
-        def _signature_src(
+        # 1) Helper: model loader (keeps existing behavior)
+        def _load_model(model_name: str, vae_model: str, task: str, controlnet_model: str) -> None:
+            # Exhaust the load_new_model generator to complete model loading
+            for _ in sd_gen.load_new_model(model_name, vae_model, task, controlnet_model):
+                pass
+
+        # 2) Thin wrapper over the existing generator pipeline
+        def _generate_image(argv: List[Any]) -> Generator[Tuple[str, Optional[List[str]], Optional[str]], None, None]:
+            # Delegate to the existing generator
+            yield from sd_gen_generate_pipeline(*argv)
+
+        # 3) Explicit-argument API: generate_image (sync)
+        def generate_image(
             # 0..6
             prompt: str,
             negative_prompt: str = "",
@@ -1952,7 +1962,7 @@ with gr.Blocks(theme=args.theme, elem_id="main", fill_width=True, fill_height=Fa
             adetailer_verbose: bool = False,
             hires_schedule_type: str = "Use same schedule type",
             hires_guidance_scale: float = -1.0,
-            # 61 controlnet model (loaderにも使用)
+            # 61 controlnet model (used in loader as well)
             controlnet_model: str = "Automatic",
             # 62..71 loop/ui/save/cache
             loop_generation: bool = False,
@@ -2011,106 +2021,209 @@ with gr.Blocks(theme=args.theme, elem_id="main", fill_width=True, fill_height=Fa
             verbose_info_gui: int = 0,
             gpu_duration: int = 20,
         ) -> Tuple[str, Optional[List[str]], Optional[str]]:
-            raise NotImplementedError
-
-        # 2) Helpers: loader + argv builder
-        def _load_model(model_name: str, vae_model: str, task: str, controlnet_model: str) -> None:
-            for _ in sd_gen.load_new_model(model_name, vae_model, task, controlnet_model):
-                pass
-
-        # Order matches sd_gen_generate_pipeline(*argv)
-        _GEN_ARG_ORDER = [
-            # keep in sync with _signature_src and pipeline
-            "prompt","negative_prompt","num_images","num_inference_steps","guidance_scale","clip_skip","seed",
-            "lora1","lora1_wt","lora2","lora2_wt","lora3","lora3_wt","lora4","lora4_wt","lora5","lora5_wt","lora6","lora6_wt","lora7","lora7_wt",
-            "sampler","schedule_type","schedule_prediction_type",
-            "height","width","model_name","vae_model","task",
-            "image_control_dict","preprocessor_name","preprocess_resolution","image_resolution",
-            "style_prompt","style_json","image_mask",
-            "strength","low_threshold","high_threshold","value_threshold","distance_threshold",
-            "recolor_gamma_correction","tile_blur_sigma",
-            "control_net_output_scaling","control_net_start_threshold","control_net_stop_threshold",
-            "textual_inversion","prompt_syntax",
-            "upscaler_model_path","upscaler_increases_size","upscaler_tile_size","upscaler_tile_overlap",
-            "hires_steps","hires_denoising_strength","hires_sampler","hires_prompt","hires_negative_prompt",
-            "adetailer_inpaint_only","adetailer_verbose","hires_schedule_type","hires_guidance_scale",
-            "controlnet_model",
-            "loop_generation","leave_progress_bar","disable_progress_bar","image_previews","display_images","save_generated_images",
-            "filename_pattern","image_storage_location",
-            "retain_compel_previous_load","retain_detailfix_model_previous_load","retain_hires_model_previous_load",
-            "t2i_adapter_preprocessor","t2i_adapter_conditioning_scale","t2i_adapter_conditioning_factor",
-            "xformers_memory_efficient_attention","free_u","generator_in_cpu",
-            "adetailer_sampler",
-            "adetailer_active_a","prompt_ad_a","negative_prompt_ad_a","strength_ad_a",
-            "face_detector_ad_a","person_detector_ad_a","hand_detector_ad_a",
-            "mask_dilation_a","mask_blur_a","mask_padding_a",
-            "adetailer_active_b","prompt_ad_b","negative_prompt_ad_b","strength_ad_b",
-            "face_detector_ad_b","person_detector_ad_b","hand_detector_ad_b",
-            "mask_dilation_b","mask_blur_b","mask_padding_b",
-            "cache_compel_texts","guidance_rescale",
-            "image_ip1_dict","mask_ip1","model_ip1","mode_ip1","scale_ip1",
-            "image_ip2_dict","mask_ip2","model_ip2","mode_ip2","scale_ip2",
-            "pag_scale","face_restoration_model","face_restoration_visibility","face_restoration_weight",
-            "load_lora_cpu","verbose_info_gui","gpu_duration",
-        ]
-
-        def _argv_from_kwargs(kwargs: Dict[str, Any]) -> List[Any]:
-            # Convert kwargs to the exact positional argv expected by generator.
-            return [kwargs.get(k) for k in _GEN_ARG_ORDER]
+            # Ensure the correct model is loaded before generation.
+            _load_model(model_name, vae_model, task, controlnet_model)
+
+            # Build argv in the exact order expected by sd_gen_generate_pipeline(*argv).
+            argv: List[Any] = [
+                # keep in sync with pipeline
+                prompt, negative_prompt, num_images, num_inference_steps, guidance_scale, clip_skip, seed,
+                lora1, lora1_wt, lora2, lora2_wt, lora3, lora3_wt, lora4, lora4_wt, lora5, lora5_wt,
+                lora6, lora6_wt, lora7, lora7_wt,
+                sampler, schedule_type, schedule_prediction_type,
+                height, width, model_name, vae_model, task,
+                image_control_dict, preprocessor_name, preprocess_resolution, image_resolution,
+                style_prompt, style_json, image_mask,
+                strength, low_threshold, high_threshold, value_threshold, distance_threshold,
+                recolor_gamma_correction, tile_blur_sigma,
+                control_net_output_scaling, control_net_start_threshold, control_net_stop_threshold,
+                textual_inversion, prompt_syntax,
+                upscaler_model_path, upscaler_increases_size, upscaler_tile_size, upscaler_tile_overlap,
+                hires_steps, hires_denoising_strength, hires_sampler, hires_prompt, hires_negative_prompt,
+                adetailer_inpaint_only, adetailer_verbose, hires_schedule_type, hires_guidance_scale,
+                controlnet_model,
+                loop_generation, leave_progress_bar, disable_progress_bar,
+                image_previews, display_images, save_generated_images,
+                filename_pattern, image_storage_location,
+                retain_compel_previous_load, retain_detailfix_model_previous_load, retain_hires_model_previous_load,
+                t2i_adapter_preprocessor, t2i_adapter_conditioning_scale, t2i_adapter_conditioning_factor,
+                xformers_memory_efficient_attention, free_u, generator_in_cpu,
+                adetailer_sampler,
+                adetailer_active_a, prompt_ad_a, negative_prompt_ad_a, strength_ad_a,
+                face_detector_ad_a, person_detector_ad_a, hand_detector_ad_a,
+                mask_dilation_a, mask_blur_a, mask_padding_a,
+                adetailer_active_b, prompt_ad_b, negative_prompt_ad_b, strength_ad_b,
+                face_detector_ad_b, person_detector_ad_b, hand_detector_ad_b,
+                mask_dilation_b, mask_blur_b, mask_padding_b,
+                cache_compel_texts, guidance_rescale,
+                image_ip1_dict, mask_ip1, model_ip1, mode_ip1, scale_ip1,
+                image_ip2_dict, mask_ip2, model_ip2, mode_ip2, scale_ip2,
+                pag_scale, face_restoration_model, face_restoration_visibility, face_restoration_weight,
+                load_lora_cpu, verbose_info_gui, gpu_duration,
+            ]
 
-        def _generate_image(argv: List[Any]) -> Generator[Tuple[str, Optional[List[str]], Optional[str]], None, None]:
-            # Delegate to existing generator
-            yield from sd_gen_generate_pipeline(*argv)
-        
-        _API_SIG = inspect.signature(_signature_src)
-
-        def _bind_api_args(*args: Any, **kwargs: Any) -> Dict[str, Any]:
-            bound = _API_SIG.bind_partial(*args, **kwargs)
-            for name, param in _API_SIG.parameters.items():
-                if name not in bound.arguments or bound.arguments[name] is None:
-                    if param.default is not inspect._empty:
-                        bound.arguments[name] = param.default
-            return bound.arguments
-
-        # 3) Signature-clone decorator: keep one signature across both endpoints
-        def clone_signature(src):
-            def deco(dst):
-                dst.__signature__ = inspect.signature(src)
-                dst.__annotations__ = src.__annotations__.copy()
-                return dst
-            return deco
-
-        # 4) Implementations: both share the same signature via clone_signature
-        @clone_signature(_signature_src)
-        def generate_image(*args: Any, **kwargs: Any) -> Tuple[str, Optional[List[str]], Optional[str]]:
-            params = _bind_api_args(*args, **kwargs)
-            _load_model(
-                params["model_name"],
-                params["vae_model"],
-                params["task"],
-                params["controlnet_model"],
-            )
-            argv = _argv_from_kwargs(params)
             last: Tuple[str, Optional[List[str]], Optional[str]] = ("COMPLETE", None, None)
             for last in _generate_image(argv):
+                # Iterate over all generator updates and keep the last tuple.
+                # This matches the behavior of returning only the final result.
                 pass
             return last
 
-        @clone_signature(_signature_src)
-        def generate_image_stream(*args: Any, **kwargs: Any) -> Generator[Tuple[str, Optional[List[str]], Optional[str]], None, None]:
-            params = _bind_api_args(*args, **kwargs)
-            _load_model(
-                params["model_name"],
-                params["vae_model"],
-                params["task"],
-                params["controlnet_model"],
-            )
-            argv = _argv_from_kwargs(params)
+        # 4) Explicit-argument API: generate_image_stream (streaming)
+        def generate_image_stream(
+            # Same signature as generate_image; kept duplicated for clarity and API docs.
+            prompt: str,
+            negative_prompt: str = "",
+            num_images: int = 1,
+            num_inference_steps: int = 28,
+            guidance_scale: float = 7.0,
+            clip_skip: int = 0,
+            seed: int = -1,
+            lora1: str = "", lora1_wt: float = 1.0,
+            lora2: str = "", lora2_wt: float = 1.0,
+            lora3: str = "", lora3_wt: float = 1.0,
+            lora4: str = "", lora4_wt: float = 1.0,
+            lora5: str = "", lora5_wt: float = 1.0,
+            lora6: str = "", lora6_wt: float = 1.0,
+            lora7: str = "", lora7_wt: float = 1.0,
+            sampler: str = "Euler",
+            schedule_type: str = "Automatic",
+            schedule_prediction_type: str = "Automatic",
+            height: int = 1024,
+            width: int = 1024,
+            model_name: str = "votepurchase/animagine-xl-3.1",
+            vae_model: str = "None",
+            task: str = "txt2img",
+            image_control_dict: Optional[dict] = None,
+            preprocessor_name: str = "Canny",
+            preprocess_resolution: int = 512,
+            image_resolution: int = 1024,
+            style_prompt: Optional[List[str]] = None,
+            style_json: Optional[dict] = None,
+            image_mask: Optional[Any] = None,
+            strength: float = 0.55,
+            low_threshold: int = 100,
+            high_threshold: int = 200,
+            value_threshold: float = 0.1,
+            distance_threshold: float = 0.1,
+            recolor_gamma_correction: float = 1.0,
+            tile_blur_sigma: int = 9,
+            control_net_output_scaling: float = 1.0,
+            control_net_start_threshold: float = 0.0,
+            control_net_stop_threshold: float = 1.0,
+            textual_inversion: bool = False,
+            prompt_syntax: str = "Classic",
+            upscaler_model_path: Optional[str] = None,
+            upscaler_increases_size: float = 1.2,
+            upscaler_tile_size: int = 0,
+            upscaler_tile_overlap: int = 8,
+            hires_steps: int = 30,
+            hires_denoising_strength: float = 0.55,
+            hires_sampler: str = "Use same sampler",
+            hires_prompt: str = "",
+            hires_negative_prompt: str = "",
+            adetailer_inpaint_only: bool = True,
+            adetailer_verbose: bool = False,
+            hires_schedule_type: str = "Use same schedule type",
+            hires_guidance_scale: float = -1.0,
+            controlnet_model: str = "Automatic",
+            loop_generation: bool = False,
+            leave_progress_bar: bool = False,
+            disable_progress_bar: bool = False,
+            image_previews: bool = True,
+            display_images: bool = True,
+            save_generated_images: bool = True,
+            filename_pattern: str = "model,seed",
+            image_storage_location: str = "./images/",
+            retain_compel_previous_load: bool = True,
+            retain_detailfix_model_previous_load: bool = True,
+            retain_hires_model_previous_load: bool = True,
+            t2i_adapter_preprocessor: Optional[str] = None,
+            t2i_adapter_conditioning_scale: float = 0.55,
+            t2i_adapter_conditioning_factor: float = 1.0,
+            xformers_memory_efficient_attention: bool = True,
+            free_u: bool = False,
+            generator_in_cpu: bool = False,
+            adetailer_sampler: str = "Use same sampler",
+            adetailer_active_a: bool = False,
+            prompt_ad_a: str = "",
+            negative_prompt_ad_a: str = "",
+            strength_ad_a: float = 0.35,
+            face_detector_ad_a: bool = False,
+            person_detector_ad_a: bool = True,
+            hand_detector_ad_a: bool = False,
+            mask_dilation_a: int = 4,
+            mask_blur_a: int = 4,
+            mask_padding_a: int = 32,
+            adetailer_active_b: bool = False,
+            prompt_ad_b: str = "",
+            negative_prompt_ad_b: str = "",
+            strength_ad_b: float = 0.35,
+            face_detector_ad_b: bool = False,
+            person_detector_ad_b: bool = True,
+            hand_detector_ad_b: bool = False,
+            mask_dilation_b: int = 4,
+            mask_blur_b: int = 4,
+            mask_padding_b: int = 32,
+            cache_compel_texts: bool = True,
+            guidance_rescale: float = 0.0,
+            image_ip1_dict: Optional[dict] = None, mask_ip1: Optional[Any] = None,
+            model_ip1: str = "plus_face", mode_ip1: str = "original", scale_ip1: float = 0.7,
+            image_ip2_dict: Optional[dict] = None, mask_ip2: Optional[Any] = None,
+            model_ip2: str = "base", mode_ip2: str = "style", scale_ip2: float = 0.7,
+            pag_scale: float = 0.0,
+            face_restoration_model: Optional[str] = None,
+            face_restoration_visibility: float = 1.0,
+            face_restoration_weight: float = 0.5,
+            load_lora_cpu: bool = False,
+            verbose_info_gui: int = 0,
+            gpu_duration: int = 20,
+        ) -> Generator[Tuple[str, Optional[List[str]], Optional[str]], None, None]:
+            # Ensure the correct model is loaded before generation.
+            _load_model(model_name, vae_model, task, controlnet_model)
+
+            argv: List[Any] = [
+                prompt, negative_prompt, num_images, num_inference_steps, guidance_scale, clip_skip, seed,
+                lora1, lora1_wt, lora2, lora2_wt, lora3, lora3_wt, lora4, lora4_wt, lora5, lora5_wt,
+                lora6, lora6_wt, lora7, lora7_wt,
+                sampler, schedule_type, schedule_prediction_type,
+                height, width, model_name, vae_model, task,
+                image_control_dict, preprocessor_name, preprocess_resolution, image_resolution,
+                style_prompt, style_json, image_mask,
+                strength, low_threshold, high_threshold, value_threshold, distance_threshold,
+                recolor_gamma_correction, tile_blur_sigma,
+                control_net_output_scaling, control_net_start_threshold, control_net_stop_threshold,
+                textual_inversion, prompt_syntax,
+                upscaler_model_path, upscaler_increases_size, upscaler_tile_size, upscaler_tile_overlap,
+                hires_steps, hires_denoising_strength, hires_sampler, hires_prompt, hires_negative_prompt,
+                adetailer_inpaint_only, adetailer_verbose, hires_schedule_type, hires_guidance_scale,
+                controlnet_model,
+                loop_generation, leave_progress_bar, disable_progress_bar,
+                image_previews, display_images, save_generated_images,
+                filename_pattern, image_storage_location,
+                retain_compel_previous_load, retain_detailfix_model_previous_load, retain_hires_model_previous_load,
+                t2i_adapter_preprocessor, t2i_adapter_conditioning_scale, t2i_adapter_conditioning_factor,
+                xformers_memory_efficient_attention, free_u, generator_in_cpu,
+                adetailer_sampler,
+                adetailer_active_a, prompt_ad_a, negative_prompt_ad_a, strength_ad_a,
+                face_detector_ad_a, person_detector_ad_a, hand_detector_ad_a,
+                mask_dilation_a, mask_blur_a, mask_padding_a,
+                adetailer_active_b, prompt_ad_b, negative_prompt_ad_b, strength_ad_b,
+                face_detector_ad_b, person_detector_ad_b, hand_detector_ad_b,
+                mask_dilation_b, mask_blur_b, mask_padding_b,
+                cache_compel_texts, guidance_rescale,
+                image_ip1_dict, mask_ip1, model_ip1, mode_ip1, scale_ip1,
+                image_ip2_dict, mask_ip2, model_ip2, mode_ip2, scale_ip2,
+                pag_scale, face_restoration_model, face_restoration_visibility, face_restoration_weight,
+                load_lora_cpu, verbose_info_gui, gpu_duration,
+            ]
+
+            # Yield all updates from the generator directly.
             yield from _generate_image(argv)
 
-        # 5) Register two APIs with identical, named kwargs
-        gr.api(generate_image, api_name="generate_image", show_api=True, queue=True, concurrency_id="gpu")
-        gr.api(generate_image_stream, api_name="generate_image_stream", show_api=True, queue=True, concurrency_id="gpu")
+    # 5) Register two APIs with explicit signatures
+    gr.api(generate_image, api_name="generate_image", show_api=True, queue=True, concurrency_id="gpu")
+    gr.api(generate_image_stream, api_name="generate_image_stream", show_api=True, queue=True, concurrency_id="gpu")
 
     gr.LoginButton()
     gr.DuplicateButton(value="Duplicate Space for private use (This demo does not work on CPU. Requires GPU Space)")