Video Agent

Video Agent is the moving-image sibling of Image Agent. Users describe a spot, watch it back, then say "make it punchier", "switch to 9:16", "try Kling" or "extend to 10 seconds" — and the agent carries the whole brief forward. Engine switching happens mid-conversation, so comparing Veo against Kling against Seedance on an identical brief is a sentence rather than a rebuild. Four aspect chips cover Reels, TikTok and Shorts alongside square, landscape and portrait, and reference images can be uploaded to anchor style or drive image-to-video.

Because video renders take minutes rather than seconds, every job is queued as a creative and finalised by the scheduled processor — no persistent queue worker required, which keeps the plugin viable on shared hosting. An optional "senior creative director" refiner expands a one-line idea into a full shot brief, with a deterministic fallback so the plugin still works when no refiner is configured. Several engines are reachable through more than one provider, so admins can use whichever account they already hold.

Key features

• Conversational video direction — follow-up turns inherit the established brief, with up to 100 turns per conversation and a 40-conversation history rail.
• Mid-thread engine switching across every video model the admin has enabled, so users can A/B vendors on the identical brief.
• Four aspect chips — 9:16 for Reels, TikTok and Shorts, plus 1:1, 16:9 and 4:5.
• Multi-provider routing per model — reach the same engine through whichever provider you have an account with.
• Cron-finalised rendering — jobs never block a web request and results appear in the gallery when ready.
• Five tweak quick-actions and four empty-state starters, so a first-time user is never staring at a blank prompt box.
• Optional creative-director LLM refiner capped at 1500 characters, with a deterministic fallback.
• Reference-image upload up to 10MB for image-to-video and style anchoring.

Powered by

Video Agent ships with four video engines, several reachable through more than one provider: Google Veo 3.1, Kling 3.0 by Kuaishou (direct, via fal.ai or via kie.ai), ByteDance Seedance 2.5 (Volcengine Ark, BytePlus ModelArk, fal.ai or kie.ai) and Runway Gen-4 Turbo. The optional prompt refiner runs on OpenAI, Google Gemini, Anthropic Claude or xAI Grok. All keys are admin-managed.

Ideal for

• Advertisers producing vertical video for Reels, TikTok and Shorts at volume.
• Teams that want to benchmark video engines against the same creative brief.
• Operators who need video generation without running a queue worker.
• Users who iterate toward a cut rather than specifying it perfectly up front.