Spaces:
Build error
Build error
Update app.py
Browse files
app.py
CHANGED
|
@@ -7,22 +7,28 @@ from PIL import Image, ImageSequence
|
|
| 7 |
from transformers import AutoProcessor, LlavaForConditionalGeneration
|
| 8 |
import gradio as gr
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
-
#
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
# Load processor and model (CPU only)
|
| 16 |
token_arg = {"token": HF_TOKEN} if HF_TOKEN else {}
|
| 17 |
processor = AutoProcessor.from_pretrained(MODEL_NAME, **token_arg)
|
| 18 |
llava_model = LlavaForConditionalGeneration.from_pretrained(
|
| 19 |
MODEL_NAME,
|
| 20 |
device_map="cpu",
|
| 21 |
torch_dtype=torch.bfloat16,
|
| 22 |
-
**token_arg
|
| 23 |
)
|
| 24 |
llava_model.eval()
|
| 25 |
|
|
|
|
|
|
|
|
|
|
| 26 |
def download_bytes(url: str, timeout: int = 30) -> bytes:
|
| 27 |
resp = requests.get(url, stream=True, timeout=timeout)
|
| 28 |
resp.raise_for_status()
|
|
@@ -39,7 +45,6 @@ def mp4_to_gif(mp4_bytes: bytes) -> bytes:
|
|
| 39 |
resp.raise_for_status()
|
| 40 |
match = re.search(r'<img[^>]+src="([^"]+\.gif)"', resp.text)
|
| 41 |
if not match:
|
| 42 |
-
# try to extract via other img tags
|
| 43 |
match = re.search(r'src="([^"]+?/tmp/[^"]+\.gif)"', resp.text)
|
| 44 |
if not match:
|
| 45 |
raise RuntimeError("Failed to extract GIF URL from ezgif response")
|
|
@@ -60,6 +65,9 @@ def load_first_frame_from_bytes(raw: bytes) -> Image.Image:
|
|
| 60 |
img = img.convert("RGB")
|
| 61 |
return img
|
| 62 |
|
|
|
|
|
|
|
|
|
|
| 63 |
def generate_caption_from_url(url: str, prompt: str = "Describe the image.") -> str:
|
| 64 |
if not url:
|
| 65 |
return "No URL provided."
|
|
@@ -70,7 +78,8 @@ def generate_caption_from_url(url: str, prompt: str = "Describe the image.") ->
|
|
| 70 |
|
| 71 |
lower = url.lower().split("?")[0]
|
| 72 |
try:
|
| 73 |
-
|
|
|
|
| 74 |
try:
|
| 75 |
raw = mp4_to_gif(raw)
|
| 76 |
except Exception as e:
|
|
@@ -89,17 +98,25 @@ def generate_caption_from_url(url: str, prompt: str = "Describe the image.") ->
|
|
| 89 |
except Exception as e:
|
| 90 |
return f"Inference error: {e}"
|
| 91 |
|
| 92 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
fn=generate_caption_from_url,
|
| 94 |
inputs=[
|
| 95 |
gr.Textbox(label="Image / GIF / MP4 URL", placeholder="https://example.com/photo.jpg"),
|
| 96 |
gr.Textbox(label="Prompt (optional)", value="Describe the image."),
|
| 97 |
],
|
| 98 |
outputs=gr.Textbox(label="Generated caption"),
|
| 99 |
-
title="JoyCaption (
|
| 100 |
description="Paste a direct link to an image, GIF, or MP4. MP4 files are converted to GIF via ezgif.com; the first frame is captioned.",
|
| 101 |
-
allow_flagging="never",
|
| 102 |
)
|
| 103 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 104 |
if __name__ == "__main__":
|
| 105 |
iface.launch()
|
|
|
|
| 7 |
from transformers import AutoProcessor, LlavaForConditionalGeneration
|
| 8 |
import gradio as gr
|
| 9 |
|
| 10 |
+
# ---------------------------
|
| 11 |
+
# Config
|
| 12 |
+
# ---------------------------
|
| 13 |
+
MODEL_NAME = "fancyfeast/llama-joycaption-beta-one-hf-llava"
|
| 14 |
+
HF_TOKEN = os.getenv("HF_TOKEN") # optional secret in Space settings
|
| 15 |
|
| 16 |
+
# ---------------------------
|
| 17 |
+
# Load model & processor
|
| 18 |
+
# ---------------------------
|
|
|
|
| 19 |
token_arg = {"token": HF_TOKEN} if HF_TOKEN else {}
|
| 20 |
processor = AutoProcessor.from_pretrained(MODEL_NAME, **token_arg)
|
| 21 |
llava_model = LlavaForConditionalGeneration.from_pretrained(
|
| 22 |
MODEL_NAME,
|
| 23 |
device_map="cpu",
|
| 24 |
torch_dtype=torch.bfloat16,
|
| 25 |
+
**token_arg,
|
| 26 |
)
|
| 27 |
llava_model.eval()
|
| 28 |
|
| 29 |
+
# ---------------------------
|
| 30 |
+
# Helpers
|
| 31 |
+
# ---------------------------
|
| 32 |
def download_bytes(url: str, timeout: int = 30) -> bytes:
|
| 33 |
resp = requests.get(url, stream=True, timeout=timeout)
|
| 34 |
resp.raise_for_status()
|
|
|
|
| 45 |
resp.raise_for_status()
|
| 46 |
match = re.search(r'<img[^>]+src="([^"]+\.gif)"', resp.text)
|
| 47 |
if not match:
|
|
|
|
| 48 |
match = re.search(r'src="([^"]+?/tmp/[^"]+\.gif)"', resp.text)
|
| 49 |
if not match:
|
| 50 |
raise RuntimeError("Failed to extract GIF URL from ezgif response")
|
|
|
|
| 65 |
img = img.convert("RGB")
|
| 66 |
return img
|
| 67 |
|
| 68 |
+
# ---------------------------
|
| 69 |
+
# Main inference
|
| 70 |
+
# ---------------------------
|
| 71 |
def generate_caption_from_url(url: str, prompt: str = "Describe the image.") -> str:
|
| 72 |
if not url:
|
| 73 |
return "No URL provided."
|
|
|
|
| 78 |
|
| 79 |
lower = url.lower().split("?")[0]
|
| 80 |
try:
|
| 81 |
+
# crude MP4 detection by extension or ftyp box signature
|
| 82 |
+
if lower.endswith(".mp4") or raw[:16].lower().find(b"ftyp") != -1:
|
| 83 |
try:
|
| 84 |
raw = mp4_to_gif(raw)
|
| 85 |
except Exception as e:
|
|
|
|
| 98 |
except Exception as e:
|
| 99 |
return f"Inference error: {e}"
|
| 100 |
|
| 101 |
+
# ---------------------------
|
| 102 |
+
# Gradio UI (compatible init)
|
| 103 |
+
# ---------------------------
|
| 104 |
+
# Use try/except to support Gradio versions that don't accept allow_flagging
|
| 105 |
+
gradio_kwargs = dict(
|
| 106 |
fn=generate_caption_from_url,
|
| 107 |
inputs=[
|
| 108 |
gr.Textbox(label="Image / GIF / MP4 URL", placeholder="https://example.com/photo.jpg"),
|
| 109 |
gr.Textbox(label="Prompt (optional)", value="Describe the image."),
|
| 110 |
],
|
| 111 |
outputs=gr.Textbox(label="Generated caption"),
|
| 112 |
+
title="JoyCaption (fancyfeast) - URL input",
|
| 113 |
description="Paste a direct link to an image, GIF, or MP4. MP4 files are converted to GIF via ezgif.com; the first frame is captioned.",
|
|
|
|
| 114 |
)
|
| 115 |
|
| 116 |
+
try:
|
| 117 |
+
iface = gr.Interface(**gradio_kwargs, allow_flagging="never")
|
| 118 |
+
except TypeError:
|
| 119 |
+
iface = gr.Interface(**gradio_kwargs)
|
| 120 |
+
|
| 121 |
if __name__ == "__main__":
|
| 122 |
iface.launch()
|