Update Notes

The hero headline now reads "A topic or script in, a long-form video in just a few clicks", putting the auto-complete-driven fast path front and centre. The old "one click" line was made honest — it now reads "One Auto Complete click finishes…" to match the real flow (topic → generate → Auto Complete). The sample-mockup "demo data" disclosure was enlarged, and the work-time figures (Auto Complete 60→5 min, render 10→2 min) are each labelled with what they actually measure.

The estimated credit cost is now shown in every input mode (topic / script / audio), with a "Covered by free credits" badge when the estimate fits the 2,000 sign-up grant. The builder now also opens on mobile — write a topic, generate, and finish with Auto Complete (fine scene-by-scene editing stays on desktop); the editor gate gained an "Auto Complete" button. Several Korean-only builder labels (shorts layout, visual-style tabs, etc.) were also moved into the localized dictionary.

Marketing / promo video creation (incl. the quick promo-video tool) is now labelled Beta. It is still being polished, so output quality may be uneven.

The 2,000-credit sign-up bonus is now granted by a DB trigger at account-creation time, not only after a completed login. Users who request a magic link but never finish signing in still receive their credits, and existing users who had no credit account were backfilled.

Structured data (JSON-LD) now matches the page language on English pages, and the Organization logo uses a proper PNG instead of the favicon. The empty "How" navigation anchor now resolves to a real section, the language toggle got an accessible label, and the generate/confirm modals now trap keyboard focus. The hero background video is skipped on phones and for reduced-motion users to keep loading light.

Hera AI and Remotion are unified under a "text-motion" provider class. Scene narration and visual ideas are compressed by GPT into a concise Korean title, 1-2 bullet phrases, a style profile (metric dashboard, kinetic typography, timeline strip, etc.), and English design/motion/graphic directions. Remotion renders locally via headless Chromium with Noto Sans KR 400/700/900 loaded by default so Korean text never falls back to tofu (□□□). Scene background images are layered as blurred underlays beneath the text.

Hera AI and Remotion prompts are now separated from normal AI-video prompts. GPT-5.4 extracts each scene's actor, event, date, and result first, then compresses only the visible screen copy into short Korean text while keeping design, motion, and graphic instructions in English. Generic filler such as "key metric", "change comparison", and "flow check" is rejected. Autocomplete can mix AI video, Hera AI, Remotion, AI image, and stock video in the same assignment order.

Remotion renders on the worker through headless Chromium. Final composition no longer applies an extra zoom motion to video clips, which reduces jitter when a Remotion/Hera/AI-video clip already contains motion. Background-image instructions for both Hera AI and Remotion were strengthened so the image remains visibly present behind the typography.

The six static AI illustrations on brief cards are gone. Each topic is matched to a Pexels stock photo through a brief-specific keyword extractor that bans abstract terms ("technology", "concept", "trend", "data", "AI") and requires concrete photographable nouns ("person using laptop", "concert stage lights"). On the Pexels response we additionally score by native resolution (DSLR-class ≥ 2400px) and penalize near-square crops to skip AI/illustration uploads. Briefs that don't match show no image rather than a generic placeholder.

Shorts-draft, queue-save, and "make today" entry points are gated behind a single flag (QUEUE_AND_DRAFT_CTAS_VISIBLE = false) across the trend page. Every "make shorts" button on hot videos / topics / news cards now reads "Make Video" and routes to /longform-builder?topic=...&desc=... instead of the hidden /shorts-draft route.

favicon.ico plus the sidebar / header / footer brand marks now share a single blue (#007AFF) base. A new scripts/rebuild-favicon.mjs renders 16/32/48/64/256 RGBA PNGs from icon.svg via sharp and writes them to both favicon.ico locations + public/brand-icon.png in one pass. Also fixes the "PNG is not in RGBA format" decode error from Rust's image crate.

Rebuilt the home page around return-visit value. Dropped the new vs. existing user split and added a top "today" digest bar (rising keywords, jobs this week, watchlist). The "Trending on YouTube right now" panel groups hot videos / surging topics / news into three tabs, with a one-click jump from any card straight into the video builder. The core feature grid is now Trends · AI Video Gen (NEW) · Image Gen · Job History · Channel Insights · Creator Guide.

Once a day an LLM analyses Korean YouTube trending data and writes 5 ready-to-make content briefs. Each brief carries a topic, "why now" rationale, hook line, three angles, recommended format (longform / shortform / both), and an estimated runtime. Cards link straight into the easy video builder. Results are cached in `daily_content_briefs` per (date, region), so the LLM bill is flat regardless of traffic.

The home Menu ① card now reads "Longform AND shortform from a script alone" and shows 16:9 longform / 9:16 shortform badges so users know both formats are supported.

The tab bar now shows only Hot Videos · Realtime Topics · News. The other tools stay in the codebase and remain reachable via direct URL (`?tab=...`), they're just hidden from the default UI.

Shorts Draft · Production Queue · Production Calendar are now hidden from both the sidebar and the home page. Trending card CTAs were renamed from "Create shorts draft" to "Make a video from this" and route to the Easy Video Maker. A new "What's new · Guide" card at the bottom of home links to update notes, the creator blog, and the How-it-works page.

Fixed contact form always failing with "submission failed". The missing source column is now auto-skipped so submissions go through, and server-side error logging/diagnostics were strengthened.

On phones the menu now hides the creation form and shows a clear "desktop-only" notice, leaving the recent-projects list for browsing progress and results. The editor page also switches to a mobile gate with project status and a deep link to the rendered video. Recent-project cards got better readability and a cleaner result-view flow.

Added per-scene video playback speed (0.25–4x) and an auto guidance hint that nudges AI images on Korean-heavy projects to render tables, charts, and callouts in natural Hangul. Refined the prompt-refiner so adjacent scenes no longer steal each other's image/video prompts.

Reworked motion, Korean info card, and ffmpeg pipeline paths to support playback speed changes, Korean text rendering, and more stable media handling.

Expanded the How-it-works page significantly and tidied the public site header/footer plus Blog, FAQ, About, and content-principles pages including their localized copy.

Added a promotions management API.

Added a looping showcase on Home and the landing page using cached 30-second lightweight previews from recent Easy Video Builder longform/shortform renders, limited to one item per format.

Improved highlight selection, changed the hook option to prepend a 3-second highlight, and clarified that captions are burned over the source video.

Stabilized uploaded SRT priority, long-video caption sync, and two-line caption display.

Future changes will be grouped by date and kept concise.