How would you optimise this agentic AI video workflow and reduce Claude usage?
retained text
Hello everyone, I'm creating an AI-generated animated series for children and use Claude Cowork as my project manager and PC operator. I'm on Pro, but regularly hit my weekly limits within a day or two. My workflow: \- Claude/Opus creates shot lists, scene prompts, timestamps and negative prompts. \- It generates stills through Higgsfield MCP. \- It controls my Windows PC, launches Pinokio and starts WanGP batch renders using Wan 2.2 Lightning. \- Video generation runs locally on my RTX 5080, sometimes producing hundreds of clips. \- Claude reviews the results using FFmpeg contact sheets, suggests fixes and repeats the process. I've already switched routine tasks to Sonnet, reduced image reviews and automated longer render batches. But the review and computer-use loops still consume a lot of usage. What would you recommend? 1. Can local scripts or models pre-screen clips so Claude only reviews problematic results? 2. Is there a more efficient way to review hundreds of generated clips? 3. Would Claude Max be worth it, or should I build a local agent using DeepSeek/Open Interpreter or another tool? 4. What's the best way to automate the PC, Pinokio and batch rendering without relying on constant computer-use sessions? I'm looking for a setup that keeps the AI in control of planning, quality decisions and corrections, while moving repetitive work locally. Any real-world experience with similar workflows would be appreciated!
document fetched 2026-11-02 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1xld6w5/how_would_you_optimise_this_agentic_ai_video