User Manual

Learn how to use AI Shorts Drama Studio to create professional short dramas, music videos, and commercials. Each stage is explained below — no prior video production experience needed. Choose Auto mode for hands-free production or Manual mode for full creative control.

How It Works
  1. Write your story idea and choose your settings (visual style, voice strategy, aspect ratio).
  2. AI generates content for the current stage (direction, script, characters, etc.).
  3. Review what AI made — read, watch, listen, or preview.
  4. Approve to continue to the next stage, or edit and regenerate.

Approval is fast, but reviewing carefully gives better final results. You do not need to finish in one sitting — your project auto-saves and resumes.

Creative Direction

The first step of every project. Type your story idea in plain language — for example, "A young couple meets at a coffee shop during a stock market crash." Or browse the Story Template Library: 360 cached story ideas across 9 genres (K-Drama, Romance, Thriller, Revenge, Social Justice, Boy Meets Girl, Time-Slip Romance, Encounter, Korean Cinderella). Click "Shuffle" to refresh, click any idea to use it. AI analyzes your idea and generates a creative direction package: title, logline, genre, emotional tone, visual style, target audience, color palette, lighting, lens suggestions, emotional arc, three-act structure, and character hints. Choices that affect the entire project: • Visual Style (10 drama styles): Photorealistic Cinema, K-Drama Cinematic, Historical Period, Pixar Disney 3D, Anime, Studio Ghibli, Spider-Verse, Noir, Cyberpunk, Webtoon. • Aspect Ratio: 16:9 Landscape for YouTube (auto-generates 9:16 portrait too). 9:16 Portrait for TikTok/Shorts (native portrait only). • Voice Strategy: Native voice (generated by video model — simplest option). ElevenLabs (Korean voice library with per-character assignment). • Voice Language: Korean, English, Japanese, or Chinese. • Production Mode: Manual (approve each stage) or Auto (hands-free with live progress). Review the generated direction, edit any field inline, then approve to continue. You can also click "Retry Direction" to regenerate from scratch.

Script

AI writes a complete screenplay based on your creative direction. The script includes scene-by-scene narrative, character dialogue, camera shot directions, and emotional pacing. Each scene has a heading (e.g., "INT. COFFEE SHOP - DAY"), action description, and character dialogue. You can review the full script, edit dialogue or scene descriptions, then approve. The script feeds into character generation, storyboarding, and voice generation.

Characters

AI creates detailed character profiles with names, ages, personality traits, physical descriptions, and wardrobe. Each character gets a unique identity that stays consistent across all generated images and video clips. Reference Images (Optional but Powerful): Upload a custom face or portrait photo for any character. When you upload a reference image, the system automatically switches to GPT Image Edit mode — instead of generating a character from scratch, it uses your uploaded image as the source and generates a multi-view character sheet that matches the face, body type, and clothing in your reference. This ensures character consistency when you have a specific look in mind. Voice Selection (depends on Voice Strategy): When voice strategy is ElevenLabs, you can browse and assign AI voices from a Korean voice library — the top 3 matching voices are recommended for each character. When voice strategy is Native, voice selection is not shown because the video model handles voice generation internally.

Character Images

AI generates multi-view character reference sheets — front view, 3/4 view, profile view, back view, and close-up headshot — for each character in your chosen visual style. These sheets ensure visual consistency across all video clips. If you uploaded a reference image, the character sheet is generated from that image using GPT Image Edit mode. If not, it is generated from the text description. You can regenerate individual character images if they do not match your vision. All images are automatically uploaded to CDN for use by the video generation models.

Storyboard

AI creates a shot-by-shot storyboard plan based on the script and characters. Each shot includes: shot type (wide, medium, close-up), camera angle, camera movement, action description, dialogue, and the scene it belongs to. The storyboard determines exactly what each video clip will contain. Large Projects (20+ shots): When a project has more than 20 estimated shots, the system automatically switches to scene-by-scene generation — instead of generating all shots in one request (which can time out or fail), it generates shots grouped by scene. If generation is interrupted, it resumes from the last completed scene. This means large dramas complete reliably without stalling or losing progress.

Voice Generation

AI generates voiceover audio for each scene's dialogue and narration. How it works depends on your Voice Strategy choice: • Native: No separate voice generation step — the video model generates voice audio built-in during video clip generation. This is the simplest option. • ElevenLabs: Uses the ElevenLabs Korean voice library. Each character speaks with their assigned AI voice. Audio is generated as separate tracks per shot, then mixed during editing. Preview the generated audio before approving. Audio quality and character voice consistency depend on this step. Commercials: Commercial projects use a silent video strategy — video clips are generated without character speech (mouth closed, no dialogue). Voiceover narration is added separately via ElevenLabs. See the Commercials section below for details.

Video Prompt Review

Before generating video clips, AI builds an optimized video prompt for each shot. These prompts include: the shot description, character bindings (e.g., "Figure 1 corresponds to Minjun"), scene environment, camera directions, dialogue, and anti-hallucination guards (preventing the AI from generating extra characters or phantom dialogue). The prompt is automatically tailored to your selected video model (Kling, Seedance, Veo, or Wan) — each model has different reference syntax and capabilities. You can review and edit any prompt before generation. This is your last chance to adjust what each shot will look like.

Video Clip Review

AI generates a video clip for each shot using the optimized prompts. Clips are generated in parallel with restartability — if a clip fails, it can be retried without regenerating the entire project. Review each clip individually. If a clip does not look right, you can regenerate it with the same or edited prompt. The regeneration uses the same character reference images and scene context. Once all clips are approved, the project moves to editing.

Editing

AI assembles all approved video clips into the final video. The editing process includes: clip concatenation in storyboard order with crossfade transitions, audio mixing with per-clip fade-in/fade-out (configurable, default 1 second each), dual-language subtitle burning (Korean + English with independent font sizes), and portrait generation. Four Output Variants (for 16:9 projects): • Landscape Subtitled — final video with burned subtitles • Landscape Unsubtitled — clean master without subtitles • Portrait Subtitled — 9:16 version for Shorts/TikTok • Portrait Unsubtitled — clean portrait master For 9:16 native projects, two variants are produced (subtitled + unsubtitled). You can adjust Korean font size (10–80), English font size (10–80), audio fade in (0–2s), and audio fade out (0–2s) before regenerating. Download all variants directly from the editing stage. Subtitles use advanced speech-to-text technology: Scribe v2 for Korean content and Forced Alignment for English songs, with quality gates for coverage and similarity. If subtitle generation fails, the video is still produced without subtitles.

Publication

AI generates platform-specific metadata for your finished video: titles, descriptions, and hashtags in both English and Korean. YouTube-specific and TikTok-specific recommendations are generated separately. Direct YouTube Upload: 1. First, connect your YouTube channel in Settings → YouTube Channels (one-time setup via Google OAuth). 2. Select your channel from the dropdown. 3. Click "Publish to YouTube" for the landscape video, or "Publish to YouTube Shorts" for the portrait/Shorts video. 4. Upload runs in the background — status is shown per video (Queued → Uploading → Published with link, or Failed). 5. The uploaded video uses AI-generated titles, descriptions, and hashtags from the publication metadata.

Music Videos

Create music videos from your own songs. Choose from 24 painting and art styles (Renaissance, Impressionism, Korean Ink, Ukiyo-e, Webtoon, and more). Upload your audio track (Suno AI or Udio), and AI generates custom artwork (40 images by default) in your chosen style, then assembles images and song into a complete music video. Subtitle Synchronization: • Korean songs: Scribe v2 speech-to-text transcribes audio and maps to lyric lines • English songs: Forced Alignment API uses known lyrics text for word-level timing — no transcription errors • Dual-language subtitles: Korean + English displayed together Unsubtitled Clean Masters: Both landscape and portrait unsubtitled copies are preserved before subtitle burning, so you can re-burn subtitles without re-composing the video. Music Video Compilation: Concatenate multiple finished music videos into a single compilation with crossfade transitions (1-second default). Videos can be ordered by duration (longest first) or by explicit index. Great for long-listening compilations. Landscape and portrait (Shorts) versions are generated. YouTube upload is available from the publication page — same channel selection and upload process as short dramas.

Commercials

Create professional commercials and product videos. The commercial pipeline has 10 stages: 1. Brief Analysis — Upload product documents (PDFs) or product URLs for AI analysis, or let AI extract a brand brief from your description. AI extracts product name, category, target audience, key features, unique selling points, and CTA. 2. Direction — AI generates a creative direction package: tone, visual style (7 commercial + 5 animation + 5 cinematic = 17 styles), segment structure, key selling points, and CTA strategy. 3. Script (VO Script) — AI writes a voiceover-driven script with per-segment breakdown (Hook, Problem, Solution, Demo, CTA). Each segment has VO text, duration, product focus, and action description. You can edit the VO script text, segments, and CTA. 4. Characters — AI creates brand spokesperson and customer persona profiles. Narrator gets voice assignment via AI-proposed voice matches. 5. Character Images — Multi-view character reference sheets in your chosen visual style. 6. Voice Generation — Per-segment VO narration using the narrator's assigned voice. You can regenerate individual segments without re-running everything. 7. Storyboard — Shot-by-shot plan with SILENT VIDEO RULES: characters do not speak on camera. Video clips contain visuals + BGM/SFX only. Voiceover is added during editing. You can edit video prompts and VO text per shot. 8. Video Generation — AI generates silent video clips. Per-clip regeneration with editable prompt and model override. 9. Editing — AI assembles all clips: clip concatenation, VO narration mixing, subtitle burning, and transitions. Landscape and portrait (Shorts) versions produced. 10. Publication — AI generates bilingual metadata (English + Korean). Direct YouTube upload for both landscape and Shorts. The 4-step creation wizard lets you choose commercial structure (Problem→Solution, Before→After, Story Driven, Product Demo, Testimonial), duration (15s/30s/60s), visual style, and AI models before starting.

Subtitle Videos

The Subtitle Video pipeline adds Korean subtitles to a video you already made outside AI-SDS. 1. Upload a Korean-language video. The source video is not regenerated, so existing footage, editing, and quality are preserved. 2. Generate Korean subtitles with speech-to-text. Review and edit the subtitle lines and timing. 3. Adjust the subtitle style and readability settings. 4. Render the subtitled video, download the result or caption files, and optionally publish it to YouTube. This is a fast, low-cost workflow for existing footage. It is separate from the creative short-drama, commercial, and music-video pipelines, which generate new video from an idea or song.

Settings — API Keys

Choose between system default API keys (managed by your administrator) or bring your own keys for each provider (AtlasCloud, Fal.ai, Higgsfield, ElevenLabs, etc.). Using your own keys gives you direct control over usage and billing — your subscription covers the workflow management, audit trail, and automated pipeline, while BYOK handles the GenAI compute costs.

Settings — YouTube Channels

Connect your YouTube channel via Google OAuth — click "Connect YouTube Channel" and authorize access. You can connect multiple channels and set one as default. Disconnect channels anytime. Connected channels are used by short drama, music video, and commercial publication stages for direct upload.

Audit

Review every AI prompt and decision at any time. The audit trail shows the exact API payload sent for each video clip — including the full prompt text, negative prompt, model name, provider used, and all parameters. This is useful for debugging, understanding AI behavior, and improving your prompts over time.

Multi-Language

The interface supports English and Korean — switch from the top navigation bar. AI-generated content (scripts, character descriptions, publication metadata) is produced in both languages. Voice generation supports Korean, English, Japanese, and Chinese.

Key Features
  • Auto-Save & Resume: You do not need to finish in one sitting. Resume whenever you want.
  • Multi-Language: Korean and English interface. AI-generated content is bilingual. Voice generation supports 4 languages.
  • Full Audit Trail: Review every AI prompt and decision at any time. Learn what works best.
  • Cinema-Grade AI: Powered by professional AI video models. High quality at zero equipment cost.
  • YouTube Publishing: Upload directly to YouTube from the publication page. Landscape and Shorts supported.
  • Reference Image Support: Upload a custom face image — AI generates character sheets consistent with your reference.
  • Scalable Generation: Scene-by-scene storyboard generation for large projects with automatic resume on failure.
  • Commercials: Create professional commercials with AI. Brand brief analysis, silent video strategy, VO narration, and per-segment voice regeneration.
  • Cost Tracking: Per-clip and per-project cost tracking in USD. See exactly how much each video costs before publishing.
  • Auto Mode: Auto mode runs the entire pipeline hands-free with live progress streaming and browser notifications.
  • Story Templates: Browse 360 story ideas across 9 genres to spark your creativity. Shuffle to refresh, click to use.
  • Unsubtitled Masters: Clean unsubtitled masters preserved for both landscape and portrait — re-burn subtitles without re-composing.
  • MV Compilation: Concatenate multiple music videos into a single compilation with crossfade transitions for long listening.