Merge pull request AUTOMATIC1111#7931 from space-nuko/img2img-enhance

Add `Upscale by` and `Upscaler` options to img2img

javascript/ui.js

@@ -132,7 +132,14 @@ function create_tab_index_args(tabId, args){
 
 function get_img2img_tab_index() {
     let res = args_to_array(arguments)
-    res.splice(-2)
+    res.splice(-2) // gradio also sends outputs to the arguments, pop them off
+    res[0] = get_tab_index('mode_img2img')
+    return res
+}
+
+function get_img2img_tab_index_for_res_preview() {
+    let res = args_to_array(arguments)
+    res.splice(-1) // gradio also sends outputs to the arguments, pop them off
     res[0] = get_tab_index('mode_img2img')
     return res
 }
@@ -361,3 +368,16 @@ function selectCheckpoint(name){
     desiredCheckpointName = name;
     gradioApp().getElementById('change_checkpoint').click()
 }
+
+
+function onCalcResolutionImg2Img(mode, scale, width, height, resize_mode, init_img, sketch, init_img_with_mask, inpaint_color_sketch, init_img_inpaint){
+    i2iScale = gradioApp().getElementById('img2img_scale')
+    i2iWidth = gradioApp().getElementById('img2img_width')
+    i2iHeight = gradioApp().getElementById('img2img_height')
+
+    setInactive(i2iScale, scale == 1)
+    setInactive(i2iWidth, scale > 1)
+    setInactive(i2iHeight, scale > 1)
+
+    return [];
+}

modules/generation_parameters_copypaste.py

@@ -282,6 +282,9 @@ def parse_generation_parameters(x: str):
         res["Hires resize-1"] = 0
         res["Hires resize-2"] = 0
 
+    if "Img2Img upscale" not in res:
+        res["Img2Img upscale"] = 1
+
     restore_old_hires_fix_params(res)
 
     return res

modules/img2img.py

@@ -78,7 +78,7 @@ def process_batch(p, input_dir, output_dir, inpaint_mask_dir, args):
                 processed_image.save(os.path.join(output_dir, filename))
 
 
-def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_styles, init_img, sketch, init_img_with_mask, inpaint_color_sketch, inpaint_color_sketch_orig, init_img_inpaint, init_mask_inpaint, steps: int, sampler_index: int, mask_blur: int, mask_alpha: float, inpainting_fill: int, restore_faces: bool, tiling: bool, n_iter: int, batch_size: int, cfg_scale: float, image_cfg_scale: float, denoising_strength: float, seed: int, subseed: int, subseed_strength: float, seed_resize_from_h: int, seed_resize_from_w: int, seed_enable_extras: bool, height: int, width: int, resize_mode: int, inpaint_full_res: bool, inpaint_full_res_padding: int, inpainting_mask_invert: int, img2img_batch_input_dir: str, img2img_batch_output_dir: str, img2img_batch_inpaint_mask_dir: str, override_settings_texts, *args):
+def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_styles, init_img, sketch, init_img_with_mask, inpaint_color_sketch, inpaint_color_sketch_orig, init_img_inpaint, init_mask_inpaint, steps: int, sampler_index: int, mask_blur: int, mask_alpha: float, inpainting_fill: int, restore_faces: bool, tiling: bool, n_iter: int, batch_size: int, cfg_scale: float, image_cfg_scale: float, denoising_strength: float, seed: int, subseed: int, subseed_strength: float, seed_resize_from_h: int, seed_resize_from_w: int, seed_enable_extras: bool, height: int, width: int, scale: float, upscaler: str, resize_mode: int, inpaint_full_res: bool, inpaint_full_res_padding: int, inpainting_mask_invert: int, img2img_batch_input_dir: str, img2img_batch_output_dir: str, img2img_batch_inpaint_mask_dir: str, override_settings_texts, *args):
     override_settings = create_override_settings_dict(override_settings_texts)
 
     is_batch = mode == 5
@@ -149,6 +149,8 @@ def img2img(id_task: str, mode: int, prompt: str, negative_prompt: str, prompt_s
         inpaint_full_res_padding=inpaint_full_res_padding,
         inpainting_mask_invert=inpainting_mask_invert,
         override_settings=override_settings,
+        scale=scale,
+        upscaler=upscaler,
     )
 
     p.scripts = modules.scripts.scripts_txt2img

modules/processing.py

@@ -946,7 +946,7 @@ def save_intermediate(image, index):
 class StableDiffusionProcessingImg2Img(StableDiffusionProcessing):
     sampler = None
 
-    def __init__(self, init_images: list = None, resize_mode: int = 0, denoising_strength: float = 0.75, image_cfg_scale: float = None, mask: Any = None, mask_blur: int = 4, inpainting_fill: int = 0, inpaint_full_res: bool = True, inpaint_full_res_padding: int = 0, inpainting_mask_invert: int = 0, initial_noise_multiplier: float = None, **kwargs):
+    def __init__(self, init_images: Optional[list] = None, resize_mode: int = 0, denoising_strength: float = 0.75, image_cfg_scale: Optional[float] = None, mask: Any = None, mask_blur: int = 4, inpainting_fill: int = 0, inpaint_full_res: bool = True, inpaint_full_res_padding: int = 0, inpainting_mask_invert: int = 0, initial_noise_multiplier: Optional[float] = None, scale: float = 0, upscaler: Optional[str] = None, **kwargs):
         super().__init__(**kwargs)
 
         self.init_images = init_images
@@ -966,11 +966,37 @@ def __init__(self, init_images: list = None, resize_mode: int = 0, denoising_str
         self.mask = None
         self.nmask = None
         self.image_conditioning = None
+        self.scale = scale
+        self.upscaler = upscaler
+
+    def get_final_size(self):
+        if self.scale > 1:
+            img = self.init_images[0]
+            width = int(img.width * self.scale)
+            height = int(img.height * self.scale)
+            return width, height
+        else:
+            return self.width, self.height
+
 
     def init(self, all_prompts, all_seeds, all_subseeds):
         self.sampler = sd_samplers.create_sampler(self.sampler_name, self.sd_model)
         crop_region = None
 
+        if self.scale > 1:
+            self.extra_generation_params["Img2Img upscale"] = self.scale
+
+        # Non-latent upscalers are run before sampling
+        # Latent upscalers are run during sampling
+        init_upscaler = None
+        if self.upscaler is not None:
+            self.extra_generation_params["Img2Img upscaler"] = self.upscaler
+            if self.upscaler not in shared.latent_upscale_modes:
+                assert len([x for x in shared.sd_upscalers if x.name == self.upscaler]) > 0, f"could not find upscaler named {self.upscaler}"
+                init_upscaler = self.upscaler
+
+        self.width, self.height = self.get_final_size()
+
         image_mask = self.image_mask
 
         if image_mask is not None:
@@ -993,7 +1019,7 @@ def init(self, all_prompts, all_seeds, all_subseeds):
                 image_mask = images.resize_image(2, mask, self.width, self.height)
                 self.paste_to = (x1, y1, x2-x1, y2-y1)
             else:
-                image_mask = images.resize_image(self.resize_mode, image_mask, self.width, self.height)
+                image_mask = images.resize_image(self.resize_mode, image_mask, self.width, self.height, init_upscaler)
                 np_mask = np.array(image_mask)
                 np_mask = np.clip((np_mask.astype(np.float32)) * 2, 0, 255).astype(np.uint8)
                 self.mask_for_overlay = Image.fromarray(np_mask)
@@ -1010,7 +1036,7 @@ def init(self, all_prompts, all_seeds, all_subseeds):
             image = images.flatten(img, opts.img2img_background_color)
 
             if crop_region is None and self.resize_mode != 3:
-                image = images.resize_image(self.resize_mode, image, self.width, self.height)
+                image = images.resize_image(self.resize_mode, image, self.width, self.height, init_upscaler)
 
             if image_mask is not None:
                 image_masked = Image.new('RGBa', (image.width, image.height))
@@ -1055,8 +1081,9 @@ def init(self, all_prompts, all_seeds, all_subseeds):
 
         self.init_latent = self.sd_model.get_first_stage_encoding(self.sd_model.encode_first_stage(image))
 
-        if self.resize_mode == 3:
-            self.init_latent = torch.nn.functional.interpolate(self.init_latent, size=(self.height // opt_f, self.width // opt_f), mode="bilinear")
+        latent_scale_mode = shared.latent_upscale_modes.get(self.upscaler, None) if self.upscaler is not None else shared.latent_upscale_modes.get(shared.latent_upscale_default_mode, "nearest")
+        if latent_scale_mode is not None:
+            self.init_latent = torch.nn.functional.interpolate(self.init_latent, size=(self.height // opt_f, self.width // opt_f), mode=latent_scale_mode["mode"], antialias=latent_scale_mode["antialias"])
 
         if image_mask is not None:
             init_mask = latent_mask

modules/ui.py

@@ -15,6 +15,7 @@
 import gradio as gr
 import gradio.routes
 import gradio.utils
+from gradio.events import Releaseable
 import numpy as np
 from PIL import Image, PngImagePlugin
 from modules.call_queue import wrap_gradio_gpu_call, wrap_queued_call, wrap_gradio_call
@@ -127,6 +128,26 @@ def calc_resolution_hires(enable, width, height, hr_scale, hr_resize_x, hr_resiz
     return f"resize: from <span class='resolution'>{p.width}x{p.height}</span> to <span class='resolution'>{p.hr_resize_x or p.hr_upscale_to_x}x{p.hr_resize_y or p.hr_upscale_to_y}</span>"
 
 
+def calc_resolution_img2img(mode, scale, resize_x, resize_y, resize_mode, *i2i_images):
+    init_img = None
+    if mode in {0, 1, 3, 4}:
+        init_img = i2i_images[mode]
+    elif mode == 2:
+        init_img = i2i_images[mode]["image"]
+
+    if not init_img:
+        return ""
+
+    if scale > 1:
+        width = int(init_img.width * scale)
+        height = int(init_img.height * scale)
+    else:
+        width = resize_x
+        height = resize_y
+
+    return f"resize: from <span class='resolution'>{init_img.width}x{init_img.height}</span> to <span class='resolution'>{width}x{height}</span>"
+
+
 def apply_styles(prompt, prompt_neg, styles):
     prompt = shared.prompt_styles.apply_styles_to_prompt(prompt, styles)
     prompt_neg = shared.prompt_styles.apply_negative_styles_to_prompt(prompt_neg, styles)
@@ -735,7 +756,7 @@ def copy_image(img):
                     )
 
                 with FormRow():
-                    resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill", "Just resize (latent upscale)"], type="index", value="Just resize")
+                    resize_mode = gr.Radio(label="Resize mode", elem_id="resize_mode", choices=["Just resize", "Crop and resize", "Resize and fill"], type="index", value="Just resize")
 
                 for category in ordered_ui_categories():
                     if category == "sampler":
@@ -744,8 +765,13 @@ def copy_image(img):
                     elif category == "dimensions":
                         with FormRow():
                             with gr.Column(elem_id="img2img_column_size", scale=4):
-                                width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width")
-                                height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height")
+                                with FormRow(variant="compact"):
+                                    final_resolution = FormHTML(value="", elem_id="img2img_finalres", label="Upscaled resolution", interactive=False)
+                                with FormRow(variant="compact"):
+                                    scale = gr.Slider(minimum=1.0, maximum=4.0, step=0.05, label="Upscale by", value=1.0, elem_id="img2img_scale")
+                                with FormRow(variant="compact"):
+                                    width = gr.Slider(minimum=64, maximum=2048, step=8, label="Width", value=512, elem_id="img2img_width")
+                                    height = gr.Slider(minimum=64, maximum=2048, step=8, label="Height", value=512, elem_id="img2img_height")
 
                             with gr.Column(elem_id="img2img_dimensions_row", scale=1, elem_classes="dimensions-tools"):
                                 res_switch_btn = ToolButton(value=switch_values_symbol, elem_id="img2img_res_switch_btn")
@@ -760,7 +786,9 @@ def copy_image(img):
                             with FormRow():
                                 cfg_scale = gr.Slider(minimum=1.0, maximum=30.0, step=0.5, label='CFG Scale', value=7.0, elem_id="img2img_cfg_scale")
                                 image_cfg_scale = gr.Slider(minimum=0, maximum=3.0, step=0.05, label='Image CFG Scale', value=1.5, elem_id="img2img_image_cfg_scale", visible=shared.sd_model and shared.sd_model.cond_stage_key == "edit")
-                            denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.75, elem_id="img2img_denoising_strength")
+                            with FormRow():
+                                upscaler = gr.Dropdown(label="Upscaler", elem_id="img2img_upscaler", choices=[*shared.latent_upscale_modes, *[x.name for x in shared.sd_upscalers]], value=shared.latent_upscale_default_mode)
+                                denoising_strength = gr.Slider(minimum=0.0, maximum=1.0, step=0.01, label='Denoising strength', value=0.75, elem_id="img2img_denoising_strength")
 
                     elif category == "seed":
                         seed, reuse_seed, subseed, reuse_subseed, subseed_strength, seed_resize_from_h, seed_resize_from_w, seed_checkbox = create_seed_inputs('img2img')
@@ -813,6 +841,39 @@ def select_img2img_tab(tab):
                                     outputs=[inpaint_controls, mask_alpha],
                                 )
 
+            img2img_resolution_preview_inputs = [dummy_component, # filled in by selected img2img tab index in _js
+                                                 scale, width, height, resize_mode,
+                                                 init_img, sketch, init_img_with_mask, inpaint_color_sketch, init_img_inpaint]
+            for input in img2img_resolution_preview_inputs[1:]:
+                if isinstance(input, Releaseable):
+                    input.release(
+                        fn=calc_resolution_img2img,
+                        _js="get_img2img_tab_index_for_res_preview",
+                        inputs=img2img_resolution_preview_inputs,
+                        outputs=[final_resolution],
+                        show_progress=False,
+                    ).success(
+                        None,
+                        _js="onCalcResolutionImg2Img",
+                        inputs=img2img_resolution_preview_inputs,
+                        outputs=[],
+                        show_progress=False,
+                    )
+                else:
+                    input.change(
+                        fn=calc_resolution_img2img,
+                        _js="get_img2img_tab_index_for_res_preview",
+                        inputs=img2img_resolution_preview_inputs,
+                        outputs=[final_resolution],
+                        show_progress=False,
+                    ).success(
+                        None,
+                        _js="onCalcResolutionImg2Img",
+                        inputs=img2img_resolution_preview_inputs,
+                        outputs=[],
+                        show_progress=False,
+                    )
+
             img2img_gallery, generation_info, html_info, html_log = create_output_panel("img2img", opts.outdir_img2img_samples)
 
             connect_reuse_seed(seed, reuse_seed, generation_info, dummy_component, is_subseed=False)
@@ -861,6 +922,8 @@ def select_img2img_tab(tab):
                     subseed, subseed_strength, seed_resize_from_h, seed_resize_from_w, seed_checkbox,
                     height,
                     width,
+                    scale,
+                    upscaler,
                     resize_mode,
                     inpaint_full_res,
                     inpaint_full_res_padding,
@@ -946,6 +1009,8 @@ def select_img2img_tab(tab):
                 (seed, "Seed"),
                 (width, "Size-1"),
                 (height, "Size-2"),
+                (scale, "Img2Img upscale"),
+                (upscaler, "Img2Img upscaler"),
                 (batch_size, "Batch size"),
                 (subseed, "Variation seed"),
                 (subseed_strength, "Variation seed strength"),

scripts/xyz_grid.py

@@ -220,6 +220,7 @@ def __init__(self, *args, **kwargs):
     AxisOption("Clip skip", int, apply_clip_skip),
     AxisOption("Denoising", float, apply_field("denoising_strength")),
     AxisOptionTxt2Img("Hires upscaler", str, apply_field("hr_upscaler"), choices=lambda: [*shared.latent_upscale_modes, *[x.name for x in shared.sd_upscalers]]),
+    AxisOptionImg2Img("Upscaler", str, apply_field("upscaler"), choices=lambda: [*shared.latent_upscale_modes, *[x.name for x in shared.sd_upscalers]]),
     AxisOptionImg2Img("Cond. Image Mask Weight", float, apply_field("inpainting_mask_weight")),
     AxisOption("VAE", str, apply_vae, cost=0.7, choices=lambda: list(sd_vae.vae_dict)),
     AxisOption("Styles", str, apply_styles, choices=lambda: list(shared.prompt_styles.styles)),

style.css

@@ -287,13 +287,13 @@ button.custom-button{
     border-radius: 0 0.5rem 0.5rem 0;
 }
 
-#txtimg_hr_finalres{
+#txtimg_hr_finalres, #img2img_finalres {
     min-height: 0 !important;
     padding: .625rem .75rem;
     margin-left: -0.75em
 }
 
-#txtimg_hr_finalres .resolution{
+#txtimg_hr_finalres .resolution, #img2img_finalres .resolution{
     font-weight: bold;
 }