Fix: save Kling Motion video to settings.paths.output_dir (matches serve endpoint)

Fix: reuse provider http client for Kling Motion, fix indentation

Fix: auto-resize character image before upload for Kling Motion

Fix: move character_orientation into function signature (was in body)

Fix: add required character_orientation param to Kling Motion API call

Fix: Kling Motion set status to 'completed' to match frontend poll check

Kling Motion: extend poll timeout to 25 min (300x5s)

Fix: correct kling-motion route path (remove double prefix)

Add Kling Motion Control: character image + driving video via WaveSpeed

Fix: add wavespeed to requirements, make SDK import optional

Sync all local changes: video routes, pod management, wavespeed, UI updates

Add LoRA strength slider, auto-set FLUX.2 defaults (CFG 2), show advanced settings for pod

Fix: request.prompt not request.positive_prompt

Fix catalog: use insert_image with correct params, store prompt info in job

Add /pod/jobs/{id}/image endpoint, show generated image directly in UI

Add pod generation progress polling with live status in UI

Increase gen timeout to 900s, add progress logging

Fix detached training launch: use setsid, close FDs, add debug logging

Add bulk caption paste feature to training UI

Fix nohup quoting: write train command to script file first

Fix coroutine .strip() bug — wrap await in parentheses

Auto-reconnect to running training pods after server restart

Run training in detached nohup process (survives SSH/server disconnect)

Fix event loop blocking: make all SSH/SFTP calls async

Add debug logging for DB path on startup

Skip local LoRA download on HF Spaces, clean up training uploads

Fix HF Spaces startup: ensure DB directory exists before engine init

Switch training UI from epochs to max steps (default 1500)

Add pod generation with FLUX.2, persistent state, training improvements

Clean up UI: smaller controls, better spacing

Make wavespeed SDK optional, use httpx fallback

Fix startup: add runpod to requirements, make import optional

Add multi-reference image support + fix missing python-dotenv

Major UI cleanup and improvements

Add detailed logging for video generation debugging

Add Higgsfield provider for Kling 3.0, Sora 2, Veo 3.1

Add video async polling + Higgsfield models via WaveSpeed

Fix async polling - check urls.get BEFORE outputs check

Fix WaveSpeed async job handling - add polling for outputs

Add WAN 2.2 and Qwen edit models to image-to-image

Fix WaveSpeed model IDs - use real API model paths

Fix: Add job tracking to img2img cloud generation

Add job tracking, status display, and cancel button

Add NSFW-friendly models to Image-to-Image edit

Add Qwen and NSFW-friendly models

Add API log panel for debugging generation requests

Add more edit models and fix img2video mode display bug

Add video model selector with cloud/pod backend support

Add password protection

Fix database directory creation at startup