Spaces:
Running
Running
Commit Β·
ee20eba
1
Parent(s): ae7d512
Remove YouTube download (blocked on HF), upload-only
Browse files- app.py +15 -53
- requirements.txt +0 -1
app.py
CHANGED
|
@@ -45,31 +45,6 @@ swapper = insightface.model_zoo.get_model(INSWAPPER_PATH, providers=PROVIDERS)
|
|
| 45 |
|
| 46 |
print("Models loaded.")
|
| 47 |
|
| 48 |
-
# ββ Helpers ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 49 |
-
def download_youtube(url: str, out_path: str):
|
| 50 |
-
raw = out_path.replace(".mp4", "_raw.mp4")
|
| 51 |
-
import yt_dlp
|
| 52 |
-
ydl_opts = {
|
| 53 |
-
"outtmpl": raw,
|
| 54 |
-
"format": "best[height<=480]/best",
|
| 55 |
-
"http_headers": {
|
| 56 |
-
"User-Agent": (
|
| 57 |
-
"Mozilla/5.0 (Windows NT 10.0; Win64; x64) "
|
| 58 |
-
"AppleWebKit/537.36 Chrome/120.0.0.0 Safari/537.36"
|
| 59 |
-
)
|
| 60 |
-
},
|
| 61 |
-
}
|
| 62 |
-
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
| 63 |
-
ydl.download([url])
|
| 64 |
-
subprocess.run(
|
| 65 |
-
["ffmpeg", "-y", "-i", raw,
|
| 66 |
-
"-vcodec", "libx264", "-acodec", "aac", "-preset", "fast",
|
| 67 |
-
out_path, "-loglevel", "error"],
|
| 68 |
-
check=True,
|
| 69 |
-
)
|
| 70 |
-
if os.path.exists(raw):
|
| 71 |
-
os.remove(raw)
|
| 72 |
-
|
| 73 |
|
| 74 |
def to_h264(src: str, dst: str):
|
| 75 |
subprocess.run(
|
|
@@ -81,11 +56,11 @@ def to_h264(src: str, dst: str):
|
|
| 81 |
|
| 82 |
|
| 83 |
# ββ Core processing ββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 84 |
-
def process(face_image, video_file,
|
| 85 |
if face_image is None:
|
| 86 |
return None, "Please upload a source face image."
|
| 87 |
-
if video_file is None
|
| 88 |
-
return None, "Please upload a video file
|
| 89 |
|
| 90 |
try:
|
| 91 |
progress(0.0, desc="Detecting source face...")
|
|
@@ -108,30 +83,25 @@ def process(face_image, video_file, youtube_url, trim_seconds, progress=gr.Progr
|
|
| 108 |
|
| 109 |
# Prepare video
|
| 110 |
progress(0.05, desc="Preparing video...")
|
| 111 |
-
raw_video
|
|
|
|
| 112 |
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
download_youtube(youtube_url.strip(), raw_video)
|
| 116 |
-
else:
|
| 117 |
-
shutil.copy(video_file, raw_video)
|
| 118 |
-
converted = f"{WORK_DIR}/temp/input_h264.mp4"
|
| 119 |
-
to_h264(raw_video, converted)
|
| 120 |
-
shutil.move(converted, raw_video)
|
| 121 |
|
| 122 |
# Verify codec
|
| 123 |
-
cap_check = cv2.VideoCapture(
|
| 124 |
ok, _ = cap_check.read()
|
| 125 |
cap_check.release()
|
| 126 |
if not ok:
|
| 127 |
-
return None, "Could not read the video
|
| 128 |
|
| 129 |
# Trim
|
| 130 |
-
input_video =
|
| 131 |
if trim_seconds and int(trim_seconds) > 0:
|
| 132 |
trimmed = f"{WORK_DIR}/temp/input_trimmed.mp4"
|
| 133 |
subprocess.run(
|
| 134 |
-
["ffmpeg", "-y", "-i",
|
| 135 |
"-t", str(int(trim_seconds)),
|
| 136 |
"-c:v", "libx264", "-c:a", "aac",
|
| 137 |
trimmed, "-loglevel", "error"],
|
|
@@ -140,7 +110,7 @@ def process(face_image, video_file, youtube_url, trim_seconds, progress=gr.Progr
|
|
| 140 |
input_video = trimmed
|
| 141 |
|
| 142 |
# Video info
|
| 143 |
-
cap
|
| 144 |
fps = cap.get(cv2.CAP_PROP_FPS)
|
| 145 |
total = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
| 146 |
w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
|
@@ -162,10 +132,8 @@ def process(face_image, video_file, youtube_url, trim_seconds, progress=gr.Progr
|
|
| 162 |
|
| 163 |
faces = face_app.get(frame)
|
| 164 |
result = frame.copy()
|
| 165 |
-
|
| 166 |
for face in faces:
|
| 167 |
result = swapper.get(result, face, source_face, paste_back=True)
|
| 168 |
-
|
| 169 |
writer.write(result)
|
| 170 |
|
| 171 |
cap.release()
|
|
@@ -185,7 +153,7 @@ def process(face_image, video_file, youtube_url, trim_seconds, progress=gr.Progr
|
|
| 185 |
|
| 186 |
progress(1.0, desc="Done!")
|
| 187 |
size = os.path.getsize(final_out) / (1024 * 1024)
|
| 188 |
-
return final_out, f"Done! {total} frames | {size:.1f} MB
|
| 189 |
|
| 190 |
except Exception as e:
|
| 191 |
return None, f"Error: {e}"
|
|
@@ -198,8 +166,7 @@ with gr.Blocks(title="Face Fusion") as demo:
|
|
| 198 |
# π Face Fusion β AI Video Face Swap
|
| 199 |
Swap any face into a video using **InsightFace + inswapper_128**.
|
| 200 |
|
| 201 |
-
> **Note:**
|
| 202 |
-
> For faster results, run the notebook on Kaggle with a free GPU.
|
| 203 |
""")
|
| 204 |
|
| 205 |
with gr.Row():
|
|
@@ -209,11 +176,6 @@ Swap any face into a video using **InsightFace + inswapper_128**.
|
|
| 209 |
type="filepath",
|
| 210 |
height=220,
|
| 211 |
)
|
| 212 |
-
youtube_input = gr.Textbox(
|
| 213 |
-
label="YouTube URL (optional)",
|
| 214 |
-
placeholder="https://www.youtube.com/watch?v=...",
|
| 215 |
-
)
|
| 216 |
-
gr.Markdown("**β or β**")
|
| 217 |
video_input = gr.Video(label="Upload Video File")
|
| 218 |
trim_input = gr.Slider(
|
| 219 |
label="Trim to first N seconds (0 = full video)",
|
|
@@ -235,7 +197,7 @@ Swap any face into a video using **InsightFace + inswapper_128**.
|
|
| 235 |
|
| 236 |
run_btn.click(
|
| 237 |
fn=process,
|
| 238 |
-
inputs=[face_input, video_input,
|
| 239 |
outputs=[video_out, status_box],
|
| 240 |
)
|
| 241 |
|
|
|
|
| 45 |
|
| 46 |
print("Models loaded.")
|
| 47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
|
| 49 |
def to_h264(src: str, dst: str):
|
| 50 |
subprocess.run(
|
|
|
|
| 56 |
|
| 57 |
|
| 58 |
# ββ Core processing ββββββββββββββββββββββββββββββββββββββββββββββββββ
|
| 59 |
+
def process(face_image, video_file, trim_seconds, progress=gr.Progress(track_tqdm=True)):
|
| 60 |
if face_image is None:
|
| 61 |
return None, "Please upload a source face image."
|
| 62 |
+
if video_file is None:
|
| 63 |
+
return None, "Please upload a video file."
|
| 64 |
|
| 65 |
try:
|
| 66 |
progress(0.0, desc="Detecting source face...")
|
|
|
|
| 83 |
|
| 84 |
# Prepare video
|
| 85 |
progress(0.05, desc="Preparing video...")
|
| 86 |
+
raw_video = f"{WORK_DIR}/temp/input.mp4"
|
| 87 |
+
converted = f"{WORK_DIR}/temp/input_h264.mp4"
|
| 88 |
|
| 89 |
+
shutil.copy(video_file, raw_video)
|
| 90 |
+
to_h264(raw_video, converted)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 91 |
|
| 92 |
# Verify codec
|
| 93 |
+
cap_check = cv2.VideoCapture(converted)
|
| 94 |
ok, _ = cap_check.read()
|
| 95 |
cap_check.release()
|
| 96 |
if not ok:
|
| 97 |
+
return None, "Could not read the video β try a different file format."
|
| 98 |
|
| 99 |
# Trim
|
| 100 |
+
input_video = converted
|
| 101 |
if trim_seconds and int(trim_seconds) > 0:
|
| 102 |
trimmed = f"{WORK_DIR}/temp/input_trimmed.mp4"
|
| 103 |
subprocess.run(
|
| 104 |
+
["ffmpeg", "-y", "-i", converted,
|
| 105 |
"-t", str(int(trim_seconds)),
|
| 106 |
"-c:v", "libx264", "-c:a", "aac",
|
| 107 |
trimmed, "-loglevel", "error"],
|
|
|
|
| 110 |
input_video = trimmed
|
| 111 |
|
| 112 |
# Video info
|
| 113 |
+
cap = cv2.VideoCapture(input_video)
|
| 114 |
fps = cap.get(cv2.CAP_PROP_FPS)
|
| 115 |
total = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
| 116 |
w = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
|
|
|
|
| 132 |
|
| 133 |
faces = face_app.get(frame)
|
| 134 |
result = frame.copy()
|
|
|
|
| 135 |
for face in faces:
|
| 136 |
result = swapper.get(result, face, source_face, paste_back=True)
|
|
|
|
| 137 |
writer.write(result)
|
| 138 |
|
| 139 |
cap.release()
|
|
|
|
| 153 |
|
| 154 |
progress(1.0, desc="Done!")
|
| 155 |
size = os.path.getsize(final_out) / (1024 * 1024)
|
| 156 |
+
return final_out, f"Done! {total} frames | {size:.1f} MB"
|
| 157 |
|
| 158 |
except Exception as e:
|
| 159 |
return None, f"Error: {e}"
|
|
|
|
| 166 |
# π Face Fusion β AI Video Face Swap
|
| 167 |
Swap any face into a video using **InsightFace + inswapper_128**.
|
| 168 |
|
| 169 |
+
> **Note:** Runs on CPU β ~1β3 min per 10 seconds of video. For GPU speed, run the notebook on Kaggle.
|
|
|
|
| 170 |
""")
|
| 171 |
|
| 172 |
with gr.Row():
|
|
|
|
| 176 |
type="filepath",
|
| 177 |
height=220,
|
| 178 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 179 |
video_input = gr.Video(label="Upload Video File")
|
| 180 |
trim_input = gr.Slider(
|
| 181 |
label="Trim to first N seconds (0 = full video)",
|
|
|
|
| 197 |
|
| 198 |
run_btn.click(
|
| 199 |
fn=process,
|
| 200 |
+
inputs=[face_input, video_input, trim_input],
|
| 201 |
outputs=[video_out, status_box],
|
| 202 |
)
|
| 203 |
|
requirements.txt
CHANGED
|
@@ -2,5 +2,4 @@ insightface==0.7.3
|
|
| 2 |
onnxruntime
|
| 3 |
opencv-python-headless
|
| 4 |
huggingface_hub
|
| 5 |
-
yt-dlp
|
| 6 |
gradio
|
|
|
|
| 2 |
onnxruntime
|
| 3 |
opencv-python-headless
|
| 4 |
huggingface_hub
|
|
|
|
| 5 |
gradio
|