Merge branch 'master' into email_zero_credits

daras_ai_v2/asr.py

@@ -384,19 +384,15 @@ def run_asr(
             if result.alternatives
         )
     elif "nemo" in selected_model.name:
-        r = requests.post(
-            str(GpuEndpoints.nemo_asr),
-            json={
-                "pipeline": dict(
-                    model_id=asr_model_ids[selected_model],
-                ),
-                "inputs": dict(
-                    audio=audio_url,
-                ),
-            },
+        data = call_celery_task(
+            "nemo_asr",
+            pipeline=dict(
+                model_id=asr_model_ids[selected_model],
+            ),
+            inputs=dict(
+                audio=audio_url,
+            ),
         )
-        r.raise_for_status()
-        data = r.json()
     # check if we should use the fast queue
     # call one of the self-hosted models
     else:

daras_ai_v2/enum_selector_widget.py

@@ -14,6 +14,12 @@ def enum_multiselect(
     checkboxes=True,
     allow_none=True,
 ):
+    try:
+        deprecated = enum_cls._deprecated()
+    except AttributeError:
+        deprecated = set()
+    enums = [e for e in enum_cls if not e in deprecated]
+
     if checkboxes:
         if label:
             st.write(label)
@@ -31,13 +37,13 @@ def render(e):
             else:
                 selected.discard(e.name)
 
-        grid_layout(2, enum_cls, render, separator=False)
+        grid_layout(2, enums, render, separator=False)
         st.session_state[key] = list(selected)
 
         return selected
     else:
         return st.multiselect(
-            options=[e.name for e in enum_cls],
+            options=[e.name for e in enums],
             format_func=lambda k: enum_cls[k].value,
             label=label,
             key=key,
@@ -53,8 +59,13 @@ def enum_selector(
     exclude: list[E] | None = None,
     **kwargs,
 ) -> str:
+    try:
+        deprecated = enum_cls._deprecated()
+    except AttributeError:
+        deprecated = set()
+    enums = [e for e in enum_cls if not e in deprecated]
     label = label or enum_cls.__name__
-    options = [e.name for e in enum_cls]
+    options = [e.name for e in enums]
     if exclude:
         options = [o for o in options if o not in exclude]
     if allow_none:

daras_ai_v2/face_restoration.py

@@ -75,11 +75,10 @@ def gfpgan(img: str, scale: int = 1) -> bytes:
     elif scale != 2:
         scale *= 2
 
-    return call_gpu_server_b64(
-        endpoint=GpuEndpoints.gfpgan,
-        input_data={
-            "img": img,
-            "version": "v1.4",
-            "scale": scale,
-        },
-    )[0]
+    # https://replicate.com/nightmareai/real-esrgan/versions/42fed1c4974146d4d2414e2be2c5277c7fcf05fcc3a73abf41610695738c1d7b#output-schema
+    model = replicate.models.get("tencentarc/gfpgan")
+    version = model.versions.get(
+        "9283608cc6b7be6b65a8e44983db012355fde4132009bf99d976b2f0896856a3"
+    )
+    img = version.predict(img=img, version="v1.4", scale=scale)
+    return requests.get(img).content

daras_ai_v2/gpu_server.py

@@ -11,34 +11,6 @@
 
 
 class GpuEndpoints:
-    wav2lip = settings.GPU_SERVER_1.copy().set(port=5001)
-    glid_3_xl_stable = settings.GPU_SERVER_1.copy().set(port=5002)
-    gfpgan = settings.GPU_SERVER_1.copy().set(port=5003)
-    dichotomous_image_segmentation = settings.GPU_SERVER_1.copy().set(port=5004)
-    # flan_t5 = f"{settings.GPU_SERVER_2}:5005"
-    # runway_ml_inpainting = f"{settings.GPU_SERVER_2}:5006"
-    u2net = settings.GPU_SERVER_1.copy().set(port=5007)
-    # deforum_sd = f"{settings.GPU_SERVER_2}:5008"
-    sd_2 = settings.GPU_SERVER_1.copy().set(port=5011)
-    # sd_multi = settings.GPU_SERVER_1.copy().set(port=5012)
-    # real_esrgan = settings.GPU_SERVER_1furl().set(port=5013)
-    # defourm_sd = settings.GPU_SERVER_2.copy().set(port=5014) / "deforum"
-
-    lavis = settings.GPU_SERVER_1.copy().set(port=5015)
-    vqa = lavis / "vqa"
-    image_captioning = lavis / "image-captioning"
-
-    _asr = settings.GPU_SERVER_1.copy().set(port=5016)
-    whisper = _asr / "whisper"
-    nemo_asr = _asr / "nemo/asr"
-
-    _asr_fast = settings.GPU_SERVER_1.copy().set(port=5019)
-    whisper_fast = _asr_fast / "whisper"
-    nemo_asr_fast = _asr_fast / "nemo/asr"
-
-    audio_ldm = settings.GPU_SERVER_1.copy().set(port=5017) / "audio_ldm"
-    bark = settings.GPU_SERVER_1.copy().set(port=5017) / "bark"
-
     deepfloyd_if = settings.GPU_SERVER_1.copy().set(port=5018) / "deepfloyd_if"
 
 

daras_ai_v2/image_segmentation.py

@@ -1,7 +1,10 @@
 from enum import Enum
 
-from daras_ai.image_input import bytes_to_cv2_img, cv2_img_to_bytes
-from daras_ai_v2.gpu_server import call_gpu_server_b64, GpuEndpoints
+import requests
+
+from daras_ai_v2.gpu_server import (
+    call_celery_task_outfile,
+)
 
 
 class ImageSegmentationModels(Enum):
@@ -10,18 +13,26 @@ class ImageSegmentationModels(Enum):
 
 
 def u2net(input_image: str) -> bytes:
-    return call_gpu_server_b64(
-        endpoint=GpuEndpoints.u2net,
-        input_data={
-            "image": input_image,
-        },
+    url = call_celery_task_outfile(
+        "u2net",
+        pipeline=dict(model_id="u2net"),
+        inputs={"images": [input_image]},
+        content_type="image/png",
+        filename="u2net.png",
     )[0]
+    r = requests.get(url)
+    r.raise_for_status()
+    return r.content
 
 
 def dis(input_image: str) -> bytes:
-    return call_gpu_server_b64(
-        endpoint=GpuEndpoints.dichotomous_image_segmentation,
-        input_data={
-            "input_image": input_image,
-        },
+    url = call_celery_task_outfile(
+        "dis",
+        pipeline=dict(model_id="isnet-general-use.pth"),
+        inputs={"images": [input_image]},
+        content_type="image/png",
+        filename="dis.png",
     )[0]
+    r = requests.get(url)
+    r.raise_for_status()
+    return r.content

daras_ai_v2/img_model_settings_widgets.py

@@ -81,7 +81,7 @@ def model_selector(
     col1, col2 = st.columns(2)
     with col1:
         selected_model = enum_selector(
-            Img2ImgModels,
+            models_enum,
             label="""
             ### 🤖 Choose your preferred AI Model
             """,
@@ -101,20 +101,20 @@ def model_selector(
         ):
             if "selected_controlnet_model" in st.session_state:
                 st.session_state["selected_controlnet_model"] = None
-        else:
+        elif models_enum is Img2ImgModels:
             enum_multiselect(
                 ControlNetModels,
                 label=controlnet_explanation,
                 key="selected_controlnet_model",
                 checkboxes=False,
                 allow_none=not require_controlnet,
             )
-        with col2:
-            controlnet_settings(
-                extra_explanations=extra_explanations,
-                low_explanation=low_explanation,
-                high_explanation=high_explanation,
-            )
+            with col2:
+                controlnet_settings(
+                    extra_explanations=extra_explanations,
+                    low_explanation=low_explanation,
+                    high_explanation=high_explanation,
+                )
     return selected_model
 
 
@@ -410,7 +410,7 @@ def prompt_strength_setting(selected_model: str = None):
 
 
 def negative_prompt_setting(selected_model: str = None):
-    if selected_model in [Text2ImgModels.dall_e.name, InpaintingModels.runway_ml.name]:
+    if selected_model in [Text2ImgModels.dall_e.name]:
         return
 
     st.text_area(

daras_ai_v2/language_model.py

@@ -54,6 +54,10 @@ class LargeLanguageModels(Enum):
 
     code_davinci_002 = "Codex [Deprecated] (openai)"
 
+    @classmethod
+    def _deprecated(cls):
+        return {cls.code_davinci_002}
+
     def is_chat_model(self) -> bool:
         return self in [
             LargeLanguageModels.gpt_4,

daras_ai_v2/settings.py

@@ -226,7 +226,6 @@
 EXPLORE_URL = furl(APP_BASE_URL).add(path="explore").url
 
 GPU_SERVER_1 = furl(config("GPU_SERVER_1", "http://gpu-1.gooey.ai"))
-GPU_SERVER_2 = furl(config("GPU_SERVER_2", "http://gpu-2.gooey.ai"))
 
 SERPER_API_KEY = config("SERPER_API_KEY", None)
 
@@ -285,3 +284,5 @@
 AZURE_FORM_RECOGNIZER_KEY = config("AZURE_FORM_RECOGNIZER_KEY", "")
 
 DEEPGRAM_API_KEY = config("DEEPGRAM_API_KEY", "")
+
+ELEVEN_LABS_API_KEY = config("ELEVEN_LABS_API_KEY", "")