🎬 dsh-videogen

AI video generation + processing for DeepSeek Harness (DSH) — turn your DSH web GUI into a video studio: multi-vendor text-to-video / image-to-video channels, FFmpeg processing (frames / GIF / compress / concat) and a prompt-driven storyboard studio, from the sidebar panel or straight from the Agent.

简体中文 | English

✨ Features

• Multi-vendor video channels in one place: OpenAI-compatible /v1/videos (Sora spec, gateway-friendly), Kling (text2video / image2video, JWT or Bearer), MiniMax Hailuo (task + file retrieve), Volcengine ARK (Doubao Seedance) and a fully custom generic async channel (submit URL, body template with {{prompt}} {{model}} {{image}} {{duration}} …, poll URL template, status/result paths) — add as many as you need
• Keys stay local: API keys live in the local DSH settings document and every upstream call is proxied by the local host; the browser and the Agent never touch plaintext credentials
• FFmpeg processing (host-side, reusing the dsh-video-tools feature set): info probe, frames extraction (evenly spaced or at a timestamp), video_to_gif (two-pass palette), image_compress, and concat with crossfade + background music
• Storyboard studio: six bundled cinematic shot templates (brand sizzle, product launch, data story, vertical social, feature orbit, release spotlight) with {{variable}} slots; generate every shot through the configured channel, then compose the whole board into one video with FFmpeg
• Agent tools: generate_video (text2video / image2video, wait or query by task_id), video_process, search_video_library, manage_storyboard — plus bundled session skills (/video:gen, /video:process, /video:storyboard)
• History + resource library: every task recorded under ~/.dsh/dsh-videogen/, generated videos cached same-origin, optional auto-save to the library with full provenance
• Prompt enhancement: rewrite a rough idea into a ready-to-generate description with the agent default model (no extra key)

📦 Installation

DSH host (Node ≥ 20) required.

dsh plugin --profile web add dsh-videogen

Local development install:

dsh plugin --profile web add /path/to/dsh-videogen

Restart dsh web — the sidebar shows the AI Video entry.

🚀 Quick start

1. Open Settings → Plugins → AI Video
2. Add a channel: pick a preset provider (+ Add provider) or + Add custom provider
3. Fill in the API URL, API key (Kling accepts AccessKey:SecretKey for JWT signing), and the model catalog — use Fetch available models to import them
4. Save, then open the AI Video sidebar panel: Generate tab (text2video / image2video, ✨ enhance), Process tab, Studio tab (templates → shots → compose) and Library tab

🎛 Channels

Preset	Submit	Poll	Auth
OpenAI-compatible	POST {base}/videos	GET /videos/{id}	Bearer
Kling	POST /v1/videos/text2video / image2video	GET …/{task_id}	Bearer token or AccessKey:SecretKey → HMAC-SHA256 JWT
MiniMax Hailuo	POST /v1/video_generation/v1 / image_to_video/v1	task poll + /v1/files/retrieve/{file_id}	Bearer (+ ?GroupId= on apiUrl)
Volcengine ARK	POST …/contents/generations/tasks	GET …/tasks/{id}	Bearer
Custom generic	your submitUrl + bodyTemplate	pollUrlTemplate	configurable header/scheme

🤖 Agent usage

Tool	Purpose
generate_video	Submit a text2video/image2video task (or query by task_id); waits by default and returns same-origin video URLs
video_process	FFmpeg processing: info / frames / gif / compress / concat on URLs or ws:<workspaceId>/<path> files
manage_storyboard	list_templates / new / get / update_shot / attach_video / compose / generate_all
search_video_library	Find and reuse previously generated/saved videos

Typical session commands (skills bundled with the plugin):

/video:gen        Cinematic shot of a neon-lit city, camera dolly-in, 9:16
/video:process    Extract 4 frames from the generated video
/video:storyboard Make a 15s product launch video for NEON X1

🔐 Security & data notes

• API keys are stored in the local DSH settings document; requests are proxied by the local host (/api/dsh-videogen/*, loopback-only routes)
• Generation consumes your upstream provider quota; video content is produced by the upstream model
• History, cached videos, processing outputs and storyboard projects persist under ~/.dsh/dsh-videogen/
• FFmpeg is resolved in order: FFMPEG_PATH → bundled @ffmpeg-installer/ffmpeg → system ffmpeg on PATH → ffmpeg-static; install via brew install ffmpeg if none exist
• Prompt enhancement calls the LLM model you chose (default: agent default model) — no extra API key

🧪 Development

pnpm install
pnpm run typecheck   # tsc --noEmit
pnpm run test        # vitest run
pnpm run build       # tsdown → lib/index.js + lib/client.js