txt2imgLoraSDXL

frontend/src/lib/components/ImagePlayer.svelte

@@ -32,5 +32,7 @@
       <slot />
     </div>
   </div>
-  <Button on:click={takeSnapshot} disabled={!isLCMRunning} classList={'ml-auto'}>Snapshot</Button>
+  <Button on:click={takeSnapshot} disabled={!isLCMRunning} classList={'text-sm my-1 ml-auto'}
+    >Snap</Button
+  >
 </div>

frontend/src/lib/components/PipelineOptions.svelte

@@ -15,31 +15,10 @@
   $: featuredOptions = pipelineParams?.filter((e) => e?.hide !== true);
 </script>
 
-<div class="grid grid-cols-1 items-center gap-3">
-  {#if featuredOptions}
-    {#each featuredOptions as params}
-      {#if params.field === FieldType.RANGE}
-        <InputRange {params} bind:value={$pipelineValues[params.id]}></InputRange>
-      {:else if params.field === FieldType.SEED}
-        <SeedInput {params} bind:value={$pipelineValues[params.id]}></SeedInput>
-      {:else if params.field === FieldType.TEXTAREA}
-        <TextArea {params} bind:value={$pipelineValues[params.id]}></TextArea>
-      {:else if params.field === FieldType.CHECKBOX}
-        <Checkbox {params} bind:value={$pipelineValues[params.id]}></Checkbox>
-      {:else if params.field === FieldType.SELECT}
-        <Selectlist {params} bind:value={$pipelineValues[params.id]}></Selectlist>
-      {/if}
-    {/each}
-  {/if}
-</div>
-
-<details>
-  <summary class="cursor-pointer font-medium">Advanced Options</summary>
-  <div
-    class="grid grid-cols-1 items-center gap-3 {pipelineParams.length > 5 ? 'sm:grid-cols-2' : ''}"
-  >
-    {#if advanceOptions}
-      {#each advanceOptions as params}
+<div class="flex flex-col gap-3">
+  <div class="grid grid-cols-1 items-center gap-3">
+    {#if featuredOptions}
+      {#each featuredOptions as params}
         {#if params.field === FieldType.RANGE}
           <InputRange {params} bind:value={$pipelineValues[params.id]}></InputRange>
         {:else if params.field === FieldType.SEED}
@@ -54,4 +33,29 @@
       {/each}
     {/if}
   </div>
-</details>
+
+  <details>
+    <summary class="cursor-pointer font-medium">Advanced Options</summary>
+    <div
+      class="grid grid-cols-1 items-center gap-3 {pipelineParams.length > 5
+        ? 'sm:grid-cols-2'
+        : ''}"
+    >
+      {#if advanceOptions}
+        {#each advanceOptions as params}
+          {#if params.field === FieldType.RANGE}
+            <InputRange {params} bind:value={$pipelineValues[params.id]}></InputRange>
+          {:else if params.field === FieldType.SEED}
+            <SeedInput {params} bind:value={$pipelineValues[params.id]}></SeedInput>
+          {:else if params.field === FieldType.TEXTAREA}
+            <TextArea {params} bind:value={$pipelineValues[params.id]}></TextArea>
+          {:else if params.field === FieldType.CHECKBOX}
+            <Checkbox {params} bind:value={$pipelineValues[params.id]}></Checkbox>
+          {:else if params.field === FieldType.SELECT}
+            <Selectlist {params} bind:value={$pipelineValues[params.id]}></Selectlist>
+          {/if}
+        {/each}
+      {/if}
+    </div>
+  </details>
+</div>

frontend/src/lib/components/TextArea.svelte

@@ -8,7 +8,7 @@
   });
 </script>
 
-<div class="px-1 py-1">
+<div class="">
   <label class="text-sm font-medium" for={params?.title}>
     {params?.title}
   </label>

frontend/src/routes/+page.svelte

@@ -111,15 +111,13 @@
     <article class="my-3 grid grid-cols-1 gap-3 lg:grid-cols-2">
       <div>
         <PipelineOptions {pipelineParams}></PipelineOptions>
-        <div class="flex gap-3">
-          <Button on:click={toggleLcmLive} {disabled}>
-            {#if isLCMRunning}
-              Stop
-            {:else}
-              Start
-            {/if}
-          </Button>
-        </div>
+        <Button on:click={toggleLcmLive} {disabled} classList={'text-lg my-1'}>
+          {#if isLCMRunning}
+            Stop
+          {:else}
+            Start
+          {/if}
+        </Button>
       </div>
       <div>
         <ImagePlayer>

pipelines/controlnetLoraSDXL.py

@@ -49,13 +49,6 @@ class Pipeline:
             field="textarea",
             id="prompt",
         )
-        model_id: str = Field(
-            "plasmo/woolitize",
-            title="Base Model",
-            values=list(base_models.keys()),
-            field="select",
-            id="model_id",
-        )
         negative_prompt: str = Field(
             default_negative_prompt,
             title="Negative Prompt",
@@ -70,10 +63,10 @@ class Pipeline:
             4, min=2, max=15, title="Steps", field="range", hide=True, id="steps"
         )
         width: int = Field(
-            512, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+            768, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
         )
         height: int = Field(
-            512, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+            768, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
         )
         guidance_scale: float = Field(
             1.0,
@@ -212,11 +205,7 @@ class Pipeline:
 
     def predict(self, params: "Pipeline.InputParams") -> Image.Image:
         generator = torch.manual_seed(params.seed)
-        print(f"Using model: {params.model_id}")
-        # pipe = self.pipes[params.model_id]
 
-        # activation_token = base_models[params.model_id]
-        # prompt = f"{activation_token} {params.prompt}"
         prompt_embeds, pooled_prompt_embeds = self.pipe.compel_proc(
             [params.prompt, params.negative_prompt]
         )

pipelines/txt2imgLoraSDXL.py

@@ -0,0 +1,142 @@
+from diffusers import (
+    DiffusionPipeline,
+    LCMScheduler,
+    AutoencoderKL,
+)
+from compel import Compel, ReturnedEmbeddingsType
+import torch
+
+try:
+    import intel_extension_for_pytorch as ipex  # type: ignore
+except:
+    pass
+
+import psutil
+from config import Args
+from pydantic import BaseModel, Field
+from PIL import Image
+
+controlnet_model = "diffusers/controlnet-canny-sdxl-1.0"
+model_id = "stabilityai/stable-diffusion-xl-base-1.0"
+lcm_lora_id = "latent-consistency/lcm-lora-sdxl"
+
+
+default_prompt = "close-up photography of old man standing in the rain at night, in a street lit by lamps, leica 35mm summilux"
+default_negative_prompt = "blurry, low quality, render, 3D, oversaturated"
+
+
+class Pipeline:
+    class Info(BaseModel):
+        name: str = "LCM+Lora+SDXL"
+        title: str = "Text-to-Image SDXL + LCM + LoRA"
+        description: str = "Generates an image from a text prompt"
+        input_mode: str = "text"
+
+    class InputParams(BaseModel):
+        prompt: str = Field(
+            default_prompt,
+            title="Prompt",
+            field="textarea",
+            id="prompt",
+        )
+        negative_prompt: str = Field(
+            default_negative_prompt,
+            title="Negative Prompt",
+            field="textarea",
+            id="negative_prompt",
+            hide=True,
+        )
+        seed: int = Field(
+            2159232, min=0, title="Seed", field="seed", hide=True, id="seed"
+        )
+        steps: int = Field(
+            4, min=2, max=15, title="Steps", field="range", hide=True, id="steps"
+        )
+        width: int = Field(
+            1024, min=2, max=15, title="Width", disabled=True, hide=True, id="width"
+        )
+        height: int = Field(
+            1024, min=2, max=15, title="Height", disabled=True, hide=True, id="height"
+        )
+        guidance_scale: float = Field(
+            1.0,
+            min=0,
+            max=20,
+            step=0.001,
+            title="Guidance Scale",
+            field="range",
+            hide=True,
+            id="guidance_scale",
+        )
+
+    def __init__(self, args: Args, device: torch.device, torch_dtype: torch.dtype):
+        vae = AutoencoderKL.from_pretrained(
+            "madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch_dtype
+        )
+        if args.safety_checker:
+            self.pipe = DiffusionPipeline.from_pretrained(
+                model_id,
+                vae=vae,
+            )
+        else:
+            self.pipe = DiffusionPipeline.from_pretrained(
+                model_id,
+                safety_checker=None,
+                vae=vae,
+            )
+        # Load LCM LoRA
+        self.pipe.load_lora_weights(lcm_lora_id, adapter_name="lcm")
+        self.pipe.scheduler = LCMScheduler.from_config(self.pipe.scheduler.config)
+        self.pipe.set_progress_bar_config(disable=True)
+        self.pipe.to(device=device, dtype=torch_dtype).to(device)
+
+        if psutil.virtual_memory().total < 64 * 1024**3:
+            self.pipe.enable_attention_slicing()
+
+        self.pipe.compel_proc = Compel(
+            tokenizer=[self.pipe.tokenizer, self.pipe.tokenizer_2],
+            text_encoder=[self.pipe.text_encoder, self.pipe.text_encoder_2],
+            returned_embeddings_type=ReturnedEmbeddingsType.PENULTIMATE_HIDDEN_STATES_NON_NORMALIZED,
+            requires_pooled=[False, True],
+        )
+
+        if args.torch_compile:
+            self.pipe.unet = torch.compile(
+                self.pipe.unet, mode="reduce-overhead", fullgraph=True
+            )
+            self.pipe.vae = torch.compile(
+                self.pipe.vae, mode="reduce-overhead", fullgraph=True
+            )
+            self.pipe(
+                prompt="warmup",
+            )
+
+    def predict(self, params: "Pipeline.InputParams") -> Image.Image:
+        generator = torch.manual_seed(params.seed)
+
+        prompt_embeds, pooled_prompt_embeds = self.pipe.compel_proc(
+            [params.prompt, params.negative_prompt]
+        )
+        results = self.pipe(
+            prompt_embeds=prompt_embeds[0:1],
+            pooled_prompt_embeds=pooled_prompt_embeds[0:1],
+            negative_prompt_embeds=prompt_embeds[1:2],
+            negative_pooled_prompt_embeds=pooled_prompt_embeds[1:2],
+            generator=generator,
+            num_inference_steps=params.steps,
+            guidance_scale=params.guidance_scale,
+            width=params.width,
+            height=params.height,
+            output_type="pil",
+        )
+
+        nsfw_content_detected = (
+            results.nsfw_content_detected[0]
+            if "nsfw_content_detected" in results
+            else False
+        )
+        if nsfw_content_detected:
+            return None
+        result_image = results.images[0]
+
+        return result_image