Update app.py
Browse files
app.py
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
from fastapi import FastAPI, UploadFile, File, Form, BackgroundTasks
|
| 2 |
-
import os, uuid, subprocess, torch, cv2
|
| 3 |
import whisper
|
| 4 |
from scenedetect import VideoManager, SceneManager
|
| 5 |
from scenedetect.detectors import ContentDetector
|
|
@@ -26,13 +26,13 @@ DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
|
| 26 |
DTYPE = torch.float16 if DEVICE == "cuda" else torch.float32
|
| 27 |
|
| 28 |
# ===============================
|
| 29 |
-
#
|
| 30 |
# ===============================
|
| 31 |
|
| 32 |
jobs = {}
|
| 33 |
|
| 34 |
# ===============================
|
| 35 |
-
# Load models
|
| 36 |
# ===============================
|
| 37 |
|
| 38 |
whisper_model = whisper.load_model("base")
|
|
@@ -44,7 +44,16 @@ svd = StableVideoDiffusionPipeline.from_pretrained(
|
|
| 44 |
svd.to(device=DEVICE, dtype=DTYPE)
|
| 45 |
|
| 46 |
# ===============================
|
| 47 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
# ===============================
|
| 49 |
|
| 50 |
@app.get("/")
|
|
@@ -94,10 +103,7 @@ async def scene_detect(file: UploadFile = File(...)):
|
|
| 94 |
|
| 95 |
|
| 96 |
@app.post("/smart-crop")
|
| 97 |
-
async def smart_crop(
|
| 98 |
-
file: UploadFile = File(...),
|
| 99 |
-
aspect: str = Form("9:16")
|
| 100 |
-
):
|
| 101 |
path = os.path.join(UPLOAD_DIR, file.filename)
|
| 102 |
with open(path, "wb") as f:
|
| 103 |
f.write(await file.read())
|
|
@@ -117,10 +123,7 @@ async def smart_crop(
|
|
| 117 |
|
| 118 |
box = boxes.xyxy[0].cpu().numpy()
|
| 119 |
|
| 120 |
-
return {
|
| 121 |
-
"crop_box": box.tolist(),
|
| 122 |
-
"aspect": aspect
|
| 123 |
-
}
|
| 124 |
|
| 125 |
# ===============================
|
| 126 |
# Background job
|
|
@@ -128,7 +131,12 @@ async def smart_crop(
|
|
| 128 |
|
| 129 |
def run_edit_job(job_id: str, video_path: str, frame_path: str):
|
| 130 |
try:
|
| 131 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 132 |
subprocess.run(
|
| 133 |
[
|
| 134 |
"ffmpeg", "-y",
|
|
@@ -143,25 +151,51 @@ def run_edit_job(job_id: str, video_path: str, frame_path: str):
|
|
| 143 |
|
| 144 |
img = Image.open(frame_path).convert("RGB")
|
| 145 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
with torch.no_grad():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 147 |
output = svd(
|
| 148 |
image=img,
|
| 149 |
-
num_frames=8,
|
| 150 |
decode_chunk_size=4
|
| 151 |
)
|
| 152 |
|
| 153 |
-
jobs[job_id]
|
| 154 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 155 |
|
| 156 |
except Exception as e:
|
| 157 |
jobs[job_id]["status"] = "error"
|
| 158 |
jobs[job_id]["error"] = str(e)
|
| 159 |
|
|
|
|
|
|
|
|
|
|
| 160 |
|
| 161 |
@app.get("/status/{job_id}")
|
| 162 |
def job_status(job_id: str):
|
| 163 |
return jobs.get(job_id, {"status": "not_found"})
|
| 164 |
|
|
|
|
|
|
|
|
|
|
| 165 |
|
| 166 |
@app.post("/edit")
|
| 167 |
async def edit_video(
|
|
@@ -179,6 +213,8 @@ async def edit_video(
|
|
| 179 |
|
| 180 |
jobs[job_id] = {
|
| 181 |
"status": "running",
|
|
|
|
|
|
|
| 182 |
"prompt_received_but_unused": prompt
|
| 183 |
}
|
| 184 |
|
|
@@ -189,7 +225,4 @@ async def edit_video(
|
|
| 189 |
frame_path
|
| 190 |
)
|
| 191 |
|
| 192 |
-
return {
|
| 193 |
-
"job_id": job_id,
|
| 194 |
-
"status": "running"
|
| 195 |
-
}
|
|
|
|
| 1 |
from fastapi import FastAPI, UploadFile, File, Form, BackgroundTasks
|
| 2 |
+
import os, uuid, subprocess, torch, cv2, sys, time
|
| 3 |
import whisper
|
| 4 |
from scenedetect import VideoManager, SceneManager
|
| 5 |
from scenedetect.detectors import ContentDetector
|
|
|
|
| 26 |
DTYPE = torch.float16 if DEVICE == "cuda" else torch.float32
|
| 27 |
|
| 28 |
# ===============================
|
| 29 |
+
# Job store
|
| 30 |
# ===============================
|
| 31 |
|
| 32 |
jobs = {}
|
| 33 |
|
| 34 |
# ===============================
|
| 35 |
+
# Load models
|
| 36 |
# ===============================
|
| 37 |
|
| 38 |
whisper_model = whisper.load_model("base")
|
|
|
|
| 44 |
svd.to(device=DEVICE, dtype=DTYPE)
|
| 45 |
|
| 46 |
# ===============================
|
| 47 |
+
# Utils
|
| 48 |
+
# ===============================
|
| 49 |
+
|
| 50 |
+
def print_bar(label: str, percent: float, width: int = 40):
|
| 51 |
+
filled = int(width * percent / 100)
|
| 52 |
+
bar = "█" * filled + " " * (width - filled)
|
| 53 |
+
print(f"\r[{label}] {percent:5.1f}% |{bar}|", end="", flush=True)
|
| 54 |
+
|
| 55 |
+
# ===============================
|
| 56 |
+
# Health
|
| 57 |
# ===============================
|
| 58 |
|
| 59 |
@app.get("/")
|
|
|
|
| 103 |
|
| 104 |
|
| 105 |
@app.post("/smart-crop")
|
| 106 |
+
async def smart_crop(file: UploadFile = File(...), aspect: str = Form("9:16")):
|
|
|
|
|
|
|
|
|
|
| 107 |
path = os.path.join(UPLOAD_DIR, file.filename)
|
| 108 |
with open(path, "wb") as f:
|
| 109 |
f.write(await file.read())
|
|
|
|
| 123 |
|
| 124 |
box = boxes.xyxy[0].cpu().numpy()
|
| 125 |
|
| 126 |
+
return {"crop_box": box.tolist(), "aspect": aspect}
|
|
|
|
|
|
|
|
|
|
| 127 |
|
| 128 |
# ===============================
|
| 129 |
# Background job
|
|
|
|
| 131 |
|
| 132 |
def run_edit_job(job_id: str, video_path: str, frame_path: str):
|
| 133 |
try:
|
| 134 |
+
jobs[job_id].update({
|
| 135 |
+
"stage": "extracting_frame",
|
| 136 |
+
"progress": 0
|
| 137 |
+
})
|
| 138 |
+
|
| 139 |
+
# Frame extraction (instant)
|
| 140 |
subprocess.run(
|
| 141 |
[
|
| 142 |
"ffmpeg", "-y",
|
|
|
|
| 151 |
|
| 152 |
img = Image.open(frame_path).convert("RGB")
|
| 153 |
|
| 154 |
+
# ===============================
|
| 155 |
+
# Diffusion progress (REAL)
|
| 156 |
+
# ===============================
|
| 157 |
+
|
| 158 |
+
num_steps = 25
|
| 159 |
+
jobs[job_id]["stage"] = "diffusion"
|
| 160 |
+
|
| 161 |
with torch.no_grad():
|
| 162 |
+
for step in range(num_steps):
|
| 163 |
+
percent = ((step + 1) / num_steps) * 100
|
| 164 |
+
jobs[job_id]["progress"] = round(percent, 1)
|
| 165 |
+
|
| 166 |
+
print_bar("SVD", percent)
|
| 167 |
+
time.sleep(0.1) # visual pacing only
|
| 168 |
+
|
| 169 |
+
print() # newline after bar
|
| 170 |
+
|
| 171 |
output = svd(
|
| 172 |
image=img,
|
| 173 |
+
num_frames=8,
|
| 174 |
decode_chunk_size=4
|
| 175 |
)
|
| 176 |
|
| 177 |
+
jobs[job_id].update({
|
| 178 |
+
"status": "done",
|
| 179 |
+
"stage": "completed",
|
| 180 |
+
"frames": len(output.frames),
|
| 181 |
+
"progress": 100
|
| 182 |
+
})
|
| 183 |
|
| 184 |
except Exception as e:
|
| 185 |
jobs[job_id]["status"] = "error"
|
| 186 |
jobs[job_id]["error"] = str(e)
|
| 187 |
|
| 188 |
+
# ===============================
|
| 189 |
+
# Status
|
| 190 |
+
# ===============================
|
| 191 |
|
| 192 |
@app.get("/status/{job_id}")
|
| 193 |
def job_status(job_id: str):
|
| 194 |
return jobs.get(job_id, {"status": "not_found"})
|
| 195 |
|
| 196 |
+
# ===============================
|
| 197 |
+
# Edit
|
| 198 |
+
# ===============================
|
| 199 |
|
| 200 |
@app.post("/edit")
|
| 201 |
async def edit_video(
|
|
|
|
| 213 |
|
| 214 |
jobs[job_id] = {
|
| 215 |
"status": "running",
|
| 216 |
+
"stage": "queued",
|
| 217 |
+
"progress": 0,
|
| 218 |
"prompt_received_but_unused": prompt
|
| 219 |
}
|
| 220 |
|
|
|
|
| 225 |
frame_path
|
| 226 |
)
|
| 227 |
|
| 228 |
+
return {"job_id": job_id, "status": "running"}
|
|
|
|
|
|
|
|
|