Qr v2

daras_ai_v2/img_model_settings_widgets.py

@@ -135,6 +135,9 @@ def controlnet_settings(
     explanations = controlnet_model_explanations | extra_explanations
 
     state_values = st.session_state.get("controlnet_conditioning_scale", [])
+    if len(models) != len(state_values):
+        state_values = state_values + [1] * len(models)
+        state_values = state_values[: len(models)]
     new_values = []
     st.write(
         """
@@ -149,17 +152,17 @@ def controlnet_settings(
         `{high_explanation.format(high=int(CONTROLNET_CONDITIONING_SCALE_RANGE[1]))}`. 
         """
     )
-    for i, model in enumerate(sorted(models)):
+    for model, value in sorted(zip(models, state_values), key=lambda x: x[0]):
         key = f"controlnet_conditioning_scale_{model}"
-        try:
-            st.session_state.setdefault(key, state_values[i])
-        except IndexError:
-            pass
+        if st.session_state.get("controlnet_overwrite"):
+            st.session_state[key] = value
+        st.session_state.setdefault(key, value)
         new_values.append(
             controlnet_weight_setting(
                 selected_controlnet_model=model, explanations=explanations, key=key
             ),
         )
+    st.session_state["selected_controlnet_model"] = sorted(models)
     st.session_state["controlnet_conditioning_scale"] = new_values
 
 
@@ -178,6 +181,9 @@ def controlnet_weight_setting(
         min_value=CONTROLNET_CONDITIONING_SCALE_RANGE[0],
         max_value=CONTROLNET_CONDITIONING_SCALE_RANGE[1],
         step=0.05,
+        default_value_attr="value"
+        if st.session_state.get("controlnet_overwrite")
+        else "defaultValue",
     )
 
 

daras_ai_v2/stable_diffusion.py

@@ -106,6 +106,7 @@ class ControlNetModels(Enum):
     sd_controlnet_seg = "Image Segmentation"
     sd_controlnet_tile = "Tiling"
     sd_controlnet_brightness = "Brightness"
+    sd_controlnet_qrmonster = "QR Monster"
 
 
 controlnet_model_explanations = {
@@ -119,6 +120,7 @@ class ControlNetModels(Enum):
     ControlNetModels.sd_controlnet_seg: "Image segmentation",
     ControlNetModels.sd_controlnet_tile: "Tiling: to preserve small details",
     ControlNetModels.sd_controlnet_brightness: "Brightness: to increase contrast naturally",
+    ControlNetModels.sd_controlnet_qrmonster: "QR Monster: make beautiful QR codes that still scan with a controlnet specifically trained for this purpose",
 }
 
 controlnet_model_ids = {
@@ -132,6 +134,7 @@ class ControlNetModels(Enum):
     ControlNetModels.sd_controlnet_seg: "lllyasviel/sd-controlnet-seg",
     ControlNetModels.sd_controlnet_tile: "lllyasviel/control_v11f1e_sd15_tile",
     ControlNetModels.sd_controlnet_brightness: "ioclab/control_v1p_sd15_brightness",
+    ControlNetModels.sd_controlnet_qrmonster: "monster-labs/control_v1p_sd15_qrcode_monster/v2",
 }
 
 
@@ -416,7 +419,7 @@ def controlnet(
     return call_sd_multi(
         "diffusion.controlnet",
         pipeline={
-            "model_id": text2img_model_ids[Text2ImgModels[selected_model]],
+            "model_id": img2img_model_ids[Img2ImgModels[selected_model]],
             "seed": seed,
             "scheduler": Schedulers[scheduler].label
             if scheduler
@@ -442,13 +445,13 @@ def controlnet(
 
 def add_prompt_prefix(prompt: str, selected_model: str) -> str:
     match selected_model:
-        case Text2ImgModels.openjourney.name:
+        case Text2ImgModels.openjourney.name | Img2ImgModels.openjourney.name:
             prompt = "mdjrny-v4 style " + prompt
-        case Text2ImgModels.analog_diffusion.name:
+        case Text2ImgModels.analog_diffusion.name | Img2ImgModels.analog_diffusion.name:
             prompt = "analog style " + prompt
-        case Text2ImgModels.protogen_5_3.name:
+        case Text2ImgModels.protogen_5_3.name | Img2ImgModels.protogen_5_3.name:
             prompt = "modelshoot style " + prompt
-        case Text2ImgModels.dreamlike_2.name:
+        case Text2ImgModels.dreamlike_2.name | Img2ImgModels.dreamlike_2.name:
             prompt = "photo, " + prompt
     return prompt
 

gooey_ui/components.py

@@ -559,6 +559,7 @@ def slider(
     help: str = None,
     *,
     disabled: bool = False,
+    default_value_attr: str = "defaultValue",
 ) -> float:
     value = _input_widget(
         input_type="range",
@@ -570,6 +571,7 @@ def slider(
         min=min_value,
         max=max_value,
         step=_step_value(min_value, max_value, step),
+        default_value_attr=default_value_attr,
     )
     return value or 0
 

recipes/QRCodeGenerator.py

@@ -1,6 +1,7 @@
 import typing
 
 import numpy as np
+from PIL import Image, ImageOps, ImageEnhance
 import qrcode
 import requests
 from django.core.exceptions import ValidationError
@@ -25,7 +26,6 @@
 )
 from daras_ai_v2.repositioning import reposition_object, repositioning_preview_widget
 from daras_ai_v2.stable_diffusion import (
-    Text2ImgModels,
     controlnet,
     ControlNetModels,
     Img2ImgModels,
@@ -35,6 +35,93 @@
 
 ATTEMPTS = 1
 
+PRESETS = {
+    "Reliable": {
+        "description": "If you just want something tried and tested, this is our original defaults.",
+        "state_update": {
+            "negative_prompt": "ugly, disfigured, low quality, blurry, nsfw, text, words",
+            "controlnet_conditioning_scale": [0.45, 0.35],
+            "guidance_scale": 9,
+            "num_outputs": 2,
+            "obj_pos_x": 0.5,
+            "obj_pos_y": 0.5,
+            "obj_scale": 0.65,
+            "output_height": 512,
+            "output_width": 512,
+            "quality": 70,
+            "scheduler": Schedulers.euler_ancestral.name,
+            "selected_controlnet_model": [
+                ControlNetModels.sd_controlnet_brightness.name,
+                ControlNetModels.sd_controlnet_tile.name,
+            ],
+            "selected_model": Img2ImgModels.dream_shaper.name,
+        },
+    },
+    "Creative": {
+        "description": "Stunning QR Codes with a creative flair that may not always be readable and could end up weird.",
+        "state_update": {
+            "negative_prompt": "ugly, disfigured, low quality, blurry, nsfw, text, words, multiple heads",
+            "controlnet_conditioning_scale": [1.4],
+            "guidance_scale": 4,
+            "num_outputs": 4,
+            "obj_pos_x": 0.5,
+            "obj_pos_y": 0.5,
+            "obj_scale": 0.65,
+            "output_height": 768,
+            "output_width": 768,
+            "quality": 40,
+            "scheduler": Schedulers.euler_ancestral.name,
+            "selected_controlnet_model": [
+                ControlNetModels.sd_controlnet_qrmonster.name,
+            ],
+            "selected_model": Img2ImgModels.dream_shaper.name,
+        },
+    },
+    "Beautiful": {
+        "description": "The best mix of reliability and creativity. Produces some of the best results for most purposes.",
+        "state_update": {
+            "negative_prompt": "ugly, disfigured, low quality, blurry, nsfw, text, words, multiple heads, many",
+            "controlnet_conditioning_scale": [0.25, 1.4],
+            "guidance_scale": 9,
+            "num_outputs": 1,
+            "obj_pos_x": 0.5,
+            "obj_pos_y": 0.5,
+            "obj_scale": 0.65,
+            "output_height": 768,
+            "output_width": 768,
+            "quality": 70,
+            "scheduler": Schedulers.euler_ancestral.name,
+            "selected_controlnet_model": [
+                ControlNetModels.sd_controlnet_brightness.name,
+                ControlNetModels.sd_controlnet_qrmonster.name,
+            ],
+            "selected_model": Img2ImgModels.dream_shaper.name,
+        },
+    },
+    "3D": {
+        "description": "Uses depth information to make the QR Code appear more 3D. This is experimental and may not always work.",
+        "state_update": {
+            "negative_prompt": "ugly, disfigured, low quality, blurry, nsfw, text, words, multiple heads, many",
+            "controlnet_conditioning_scale": [0.35, 0.3, 0.3],
+            "guidance_scale": 8,
+            "num_outputs": 1,
+            "obj_pos_x": 0.5,
+            "obj_pos_y": 0.5,
+            "obj_scale": 0.8,
+            "output_height": 512,
+            "output_width": 512,
+            "quality": 100,
+            "scheduler": Schedulers.euler_ancestral.name,
+            "selected_controlnet_model": [
+                ControlNetModels.sd_controlnet_brightness.name,
+                ControlNetModels.sd_controlnet_depth.name,
+                ControlNetModels.sd_controlnet_tile.name,
+            ],
+            "selected_model": Img2ImgModels.dream_shaper.name,
+        },
+    },
+}
+
 
 class QRCodeGeneratorPage(BasePage):
     title = "AI Art QR Code"
@@ -45,6 +132,7 @@ class QRCodeGeneratorPage(BasePage):
         obj_scale=0.65,
         obj_pos_x=0.5,
         obj_pos_y=0.5,
+        color=255,
     )
 
     def __init__(self, *args, **kwargs):
@@ -60,7 +148,7 @@ class RequestModel(BaseModel):
         text_prompt: str
         negative_prompt: str | None
 
-        selected_model: typing.Literal[tuple(e.name for e in Text2ImgModels)] | None
+        selected_model: typing.Literal[tuple(e.name for e in Img2ImgModels)] | None
         selected_controlnet_model: list[
             typing.Literal[tuple(e.name for e in ControlNetModels)], ...
         ] | None
@@ -80,6 +168,7 @@ class RequestModel(BaseModel):
         obj_scale: float | None
         obj_pos_x: float | None
         obj_pos_y: float | None
+        color: int | None
 
     class ResponseModel(BaseModel):
         output_images: list[str]
@@ -212,6 +301,24 @@ def render_settings(self):
             """
         )
 
+        preset_match = None
+        st.session_state["controlnet_overwrite"] = False
+        for preset, preset_value in PRESETS.items():
+            key = "preset_" + preset
+            if st.button(preset, key=key, disabled=st.session_state.get(key, False)):
+                preset_match = preset_value
+                st.session_state.update(preset_value["state_update"])
+                st.session_state["controlnet_overwrite"] = True
+        if not preset_match:
+            st.button("Custom", key="preset_Custom", disabled=True)
+            st.caption(
+                "For the tech savvy and the curious. Here you can play around with settings to create something truly unique."
+            )
+        else:
+            st.caption(
+                preset_match["description"],
+            )
+
         img_model_settings(
             Img2ImgModels,
             show_scheduler=True,
@@ -265,6 +372,9 @@ def render_settings(self):
         img_cv2 = mask_cv2 = np.array(
             qrcode.QRCode(border=0).make_image().convert("RGB")
         )
+        color = st.slider(
+            "`Grayscale background`", min_value=0, max_value=255, key="color"
+        )
         repositioning_preview_widget(
             img_cv2=img_cv2,
             mask_cv2=mask_cv2,
@@ -275,7 +385,7 @@ def render_settings(self):
                 st.session_state["output_width"],
                 st.session_state["output_height"],
             ),
-            color=255,
+            color=color,
         )
 
     def render_output(self):
@@ -328,7 +438,7 @@ def run(self, state: dict) -> typing.Iterator[str | None]:
 
         state["raw_images"] = raw_images = []
 
-        yield f"Running {Text2ImgModels[request.selected_model].value}..."
+        yield f"Running {Img2ImgModels[request.selected_model].value}..."
         state["output_images"] = controlnet(
             selected_model=request.selected_model,
             selected_controlnet_model=request.selected_controlnet_model,
@@ -364,12 +474,10 @@ def preview_description(self, state: dict) -> str:
         """
 
     def get_raw_price(self, state: dict) -> int:
-        selected_model = state.get("selected_model", Text2ImgModels.dream_shaper.name)
-        total = 5
+        selected_model = state.get("selected_model", Img2ImgModels.dream_shaper.name)
+        total = 30
         match selected_model:
-            case Text2ImgModels.deepfloyd_if.name:
-                total += 3
-            case Text2ImgModels.dall_e.name:
+            case Img2ImgModels.dall_e.name:
                 total += 10
         return total * state.get("num_outputs", 1)
 
@@ -412,7 +520,7 @@ def generate_and_upload_qr_code(
         out_obj_scale=request.obj_scale,
         out_pos_x=request.obj_pos_x,
         out_pos_y=request.obj_pos_y,
-        color=255,
+        color=request.color,
     )
 
     img_url = upload_file_from_bytes("cleaned_qr.png", cv2_img_to_bytes(img_cv2))
@@ -432,14 +540,40 @@ def download_qr_code_data(url: str) -> str:
     return extract_qr_code_data(img)
 
 
-def extract_qr_code_data(img: np.ndarray) -> str:
-    decoded = pyzbar.decode(img)
-    if not (decoded and decoded[0]):
+def extract_qr_code_data(image: np.ndarray) -> str:
+    # cycle through different sizes and sharpnesses, etc. for a more robust qr code extraction
+    image = Image.fromarray(image)
+    found_qr_code = False
+    found_qr_code_data = False
+    x, y = image.size
+    for scalar in [0.1, 0.2, 0.5, 1]:
+        image_scaled = image.resize((int(round(x * scalar)), int(round(y * scalar))))
+        for sharpness in [0.1, 0.5, 1, 2]:
+            image_scaled_sharp = ImageEnhance.Sharpness(image_scaled).enhance(sharpness)
+            image_autocontrast = ImageOps.autocontrast(image_scaled_sharp)
+            image_inverted = ImageOps.invert(image_autocontrast)
+            image_grayscale = ImageOps.grayscale(image_autocontrast)
+            image_grayscale_inverted = ImageOps.grayscale(image_inverted)
+            for img in [
+                image_autocontrast,
+                image_inverted,
+                image_grayscale,
+                image_grayscale_inverted,
+            ]:
+                decoded = pyzbar.decode(img)
+                if decoded and decoded[0]:
+                    found_qr_code = True
+                    info = decoded[0].data.decode()
+                    if info:
+                        found_qr_code_data = True
+                        print(f"QR code data: {info}")
+                        return info
+
+    if not found_qr_code:
         raise InvalidQRCode("No QR code found in image")
-    info = decoded[0].data.decode()
-    if not info:
+
+    if not found_qr_code_data:
         raise InvalidQRCode("No data found in QR code")
-    return info
 
 
 class InvalidQRCode(AssertionError):