upgrade callback timeout to 600

fix: VRAM monitory for server and soft lazy loading

change: to safe flow model by presure

add: @asynccontextmanager

fix: leak memory from models

tasks_in_queue = max(0, tasks_in_queue - 1)

add: queue for customers

fix: prevent concurrent model access using global lock

remove: all narrative flow and flash_attn

fix: cpu and missed AudioFileClip

feature: audio AI voice for per image

remove: Optional[NarrationGenerator]

feature: OpenAI text speech for images

add: calback video by webhook

refactor: app and logic with right upload TTS

fix: added os import

fix: UPPER to upper

fix: app start with remover func

test: added new model for generate video with AI voices

fix: increase requests.post timeout from 60 to 180 seconds

fix: extra parantesses

feat: resize image to 90 procent

revert: to the 39b31b5 v4

fix: rename logicV3 to logicV4

add: logicV4.py

add: WatermarkRemoverV3

refactor: upgrade detection logic to logicV2 with wavelet support

revert: to the 39b31b5 v3

fix: big size for images

revert: to Lama model v2 - 39b31b5

revert: to 8fded30 version

fix: resolve torch reference with explicit in-method import

fix: resolve torch import issue in lazy-loaded module

feat: add xformers support with fallback and GPU memory optimization

feat: increase expand_percentage to quality

add: jpeg for all iterations

feat(app): Reduce output file size by saving as JPEG

fix(app): Restore logger definition

refactor(app): Defer heavy imports to improve startup time

fix: all requirements and docker build

fix: Overhaul model loading for stability & memory efficiency

fix: infinity restart for HF

fix: move to greedy upload models

revert: to Diffusion model for test

revert: to Florence version v2

Fix: Use string forward references for type hints

Refactor: Defer 'logic' import to prevent startup crash

Refactor: Implement lazy imports to prevent startup crash

improve: State-of-the-Art

feat: Implement lazy loading for models