Update app.py

app.py

@@ -1,154 +1,76 @@
-import gradio as gr
-import numpy as np
-import random
+import os
 
-# import spaces #[uncomment to use ZeroGPU]
-from diffusers import DiffusionPipeline
+import spaces
 import torch
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
-
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
-
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
-
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
-
-# @spaces.GPU #[uncomment to use ZeroGPU]
-def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    progress=gr.Progress(track_tqdm=True),
-):
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-
-    generator = torch.Generator().manual_seed(seed)
-
+import gradio as gr
+from diffusers import Ideogram4Pipeline
+
+# Use the nf4-quantized repo -- this is the one diffusers' Ideogram4Pipeline
+# actually supports. The -fp8 repo ships its own custom runtime and is not
+# loadable via from_pretrained.
+MODEL_ID = "ideogram-ai/ideogram-4-nf4"
+
+# ZeroGPU Spaces need a Space secret (Settings -> Variables and secrets) named
+# HF_TOKEN, since Ideogram 4's weights are gated and require an authenticated,
+# license-accepted account to download.
+HF_TOKEN = os.environ.get("HF_TOKEN")
+
+# Load once at import time, on CPU. ZeroGPU intercepts `.to("cuda")` here so
+# this is safe even though no GPU is attached to the process yet -- the real
+# GPU is only attached for the duration of a call into an @spaces.GPU function.
+pipe = Ideogram4Pipeline.from_pretrained(
+    MODEL_ID,
+    dtype=torch.bfloat16,
+    token=HF_TOKEN,
+).to("cuda")
+
+
+@spaces.GPU(duration=120)  # seconds of GPU time granted per call; raise if you increase steps/resolution
+def generate(prompt, negative_prompt, width, height, guidance_scale, steps, seed):
+    generator = torch.Generator("cuda").manual_seed(int(seed))
     image = pipe(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
-        width=width,
-        height=height,
+        prompt,
+        negative_prompt=negative_prompt or None,
+        height=int(height),
+        width=int(width),
+        guidance_scale=float(guidance_scale),
+        num_inference_steps=int(steps),
         generator=generator,
     ).images[0]
+    return image
 
-    return image, seed
-
-
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css = """
-#col-container {
-    margin: 0 auto;
-    max-width: 640px;
-}
-"""
-
-with gr.Blocks(css=css) as demo:
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(" # Text-to-Image Gradio Template")
 
-        with gr.Row():
-            prompt = gr.Text(
+with gr.Blocks(title="Ideogram 4") as demo:
+    gr.Markdown(
+        "# Ideogram 4\n"
+        "Text-to-image with [Ideogram 4](https://huggingface.co/ideogram-ai/ideogram-4-nf4), "
+        "running on ZeroGPU. Note: Ideogram 4 was trained on structured JSON captions -- "
+        "short natural-language prompts still work, but for best results expand your "
+        "prompt into their JSON schema first (see the model card)."
+    )
+    with gr.Row():
+        with gr.Column():
+            prompt = gr.Textbox(
                 label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-
-            run_button = gr.Button("Run", scale=0, variant="primary")
-
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
+                lines=3,
+                placeholder="A photo of a cat holding a sign that says hello world",
             )
-
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-
+            negative_prompt = gr.Textbox(label="Negative prompt (optional)", lines=2)
             with gr.Row():
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
+                width = gr.Slider(512, 2048, value=1024, step=64, label="Width")
+                height = gr.Slider(512, 2048, value=1024, step=64, label="Height")
             with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
-                )
-
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2,  # Replace with defaults that work for your model
-                )
-
-        gr.Examples(examples=examples, inputs=[prompt])
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
+                guidance_scale = gr.Slider(1.0, 15.0, value=7.0, step=0.5, label="Guidance scale")
+                steps = gr.Slider(4, 60, value=48, step=1, label="Steps")
+            seed = gr.Number(value=0, label="Seed", precision=0)
+            run_btn = gr.Button("Generate", variant="primary")
+        with gr.Column():
+            output = gr.Image(label="Result")
+
+    run_btn.click(
+        fn=generate,
+        inputs=[prompt, negative_prompt, width, height, guidance_scale, steps, seed],
+        outputs=output,
     )
 
-if __name__ == "__main__":
-    demo.launch()
+demo.queue().launch()