Update app.py
Browse files
app.py
CHANGED
@@ -4,25 +4,21 @@ from diffusers import StableDiffusionPipeline
|
|
4 |
import torch
|
5 |
import os
|
6 |
|
7 |
-
# 1. Check for HF_TOKEN
|
8 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
9 |
if HF_TOKEN is None:
|
10 |
-
raise ValueError("
|
11 |
|
12 |
-
# 2. Set device
|
13 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
14 |
|
15 |
-
#
|
16 |
translator = pipeline(
|
17 |
"translation",
|
18 |
-
model="Helsinki-NLP/opus-mt-
|
19 |
use_auth_token=HF_TOKEN
|
20 |
)
|
21 |
|
22 |
-
# 4. Load text generator (GPT-2) — public, no token needed
|
23 |
generator = pipeline("text-generation", model="gpt2")
|
24 |
|
25 |
-
# 5. Load image generator (Stable Diffusion) with token
|
26 |
image_pipe = StableDiffusionPipeline.from_pretrained(
|
27 |
"CompVis/stable-diffusion-v1-4",
|
28 |
use_auth_token=HF_TOKEN,
|
@@ -30,33 +26,21 @@ image_pipe = StableDiffusionPipeline.from_pretrained(
|
|
30 |
)
|
31 |
image_pipe = image_pipe.to(device)
|
32 |
|
33 |
-
# 6. Main function
|
34 |
def generate_image_from_tamil(tamil_input):
|
35 |
-
# Translate Tamil to English
|
36 |
translated = translator(tamil_input, max_length=100)[0]['translation_text']
|
37 |
-
|
38 |
-
# Generate a prompt using GPT-2
|
39 |
-
generated = generator(translated, max_length=50, num_return_sequences=1)[0]['generated_text']
|
40 |
-
generated = generated.strip()
|
41 |
-
|
42 |
-
# Generate image using Stable Diffusion
|
43 |
image = image_pipe(generated).images[0]
|
44 |
-
|
45 |
return translated, generated, image
|
46 |
|
47 |
-
# 7. Gradio Interface
|
48 |
iface = gr.Interface(
|
49 |
fn=generate_image_from_tamil,
|
50 |
inputs=gr.Textbox(lines=2, label="Enter Tamil Text"),
|
51 |
-
outputs=[
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
title="Tamil to Image Generator",
|
57 |
-
description="This app translates Tamil text to English, generates creative English prompts, and visualizes them using Stable Diffusion.",
|
58 |
allow_flagging="never"
|
59 |
)
|
60 |
|
61 |
-
# 8. Launch app
|
62 |
iface.launch()
|
|
|
4 |
import torch
|
5 |
import os
|
6 |
|
|
|
7 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
8 |
if HF_TOKEN is None:
|
9 |
+
raise ValueError("Set HF_TOKEN in env variables.")
|
10 |
|
|
|
11 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
12 |
|
13 |
+
# ✅ Use multilingual model that supports Tamil→English
|
14 |
translator = pipeline(
|
15 |
"translation",
|
16 |
+
model="Helsinki-NLP/opus-mt-mul-en",
|
17 |
use_auth_token=HF_TOKEN
|
18 |
)
|
19 |
|
|
|
20 |
generator = pipeline("text-generation", model="gpt2")
|
21 |
|
|
|
22 |
image_pipe = StableDiffusionPipeline.from_pretrained(
|
23 |
"CompVis/stable-diffusion-v1-4",
|
24 |
use_auth_token=HF_TOKEN,
|
|
|
26 |
)
|
27 |
image_pipe = image_pipe.to(device)
|
28 |
|
|
|
29 |
def generate_image_from_tamil(tamil_input):
|
|
|
30 |
translated = translator(tamil_input, max_length=100)[0]['translation_text']
|
31 |
+
generated = generator(translated, max_length=50, num_return_sequences=1)[0]['generated_text'].strip()
|
|
|
|
|
|
|
|
|
|
|
32 |
image = image_pipe(generated).images[0]
|
|
|
33 |
return translated, generated, image
|
34 |
|
|
|
35 |
iface = gr.Interface(
|
36 |
fn=generate_image_from_tamil,
|
37 |
inputs=gr.Textbox(lines=2, label="Enter Tamil Text"),
|
38 |
+
outputs=[gr.Textbox(label="Translated English Text"),
|
39 |
+
gr.Textbox(label="Generated English Prompt"),
|
40 |
+
gr.Image(label="Generated Image")],
|
41 |
+
title="Tamil→Image Generator",
|
42 |
+
description="Translate Tamil → English, generate prompt → create image.",
|
|
|
|
|
43 |
allow_flagging="never"
|
44 |
)
|
45 |
|
|
|
46 |
iface.launch()
|