Agent-driven AI Video
Direct video like you code
VFXLY is an agent-driven workspace for AI video: plan scenes, pick your models and render — from one prompt to a full edit, with the same rhythm as shipping software.
Launching soon
VFXLY opens September 1, 2026
US West Coast, 00:00 PT. Be among the first to direct AI video like a pro.
Everything creators need
From first idea to final take — one tool, one workflow.
Projects & tabs
Like a terminal for content: every project has its own tabs (sessions) and its own agent playbook. Render, compare and iterate in parallel.
Agent loop
Describe a video — the agent plans a storyboard (scenes, script, shot, camera, aspect), renders every scene and flags weak takes for regeneration.
Multi-provider
One adapter layer for Runway, Pika, Kling and Seedance. Jobs with polling, unified result assets in the gallery.
Agent-driven workflows
Describe complex pipelines as Markdown: style, rules, character sheets, shot lists. The agent executes the instructions step by step.
MCP on strong servers
OpenRouter drives the brain, MCP servers drive open-source models (ComfyUI & co.) on powerful hardware.
Iteration & gallery
All takes stored centrally. Thumbnails, comparisons, prompts & parameters per generation — for real creative control.
From prompt to take
Plan
Describe your goal. The agent loads the project's playbook and builds a storyboard with scenes, script, camera & aspect ratio.
Execute
Every scene runs through the matching provider adapter. Jobs with polling, retry and a unified asset format.
Iterate
Compare takes in the gallery, regenerate weak results, fine-tune parameters — until the take is right.
One interface, many models
Be first in line
VFXLY launches September 1, 2026. Join the waitlist to get early access and launch announcements.