Automated AI 2K documentaries in ComfyUI: Qwen Image 2.1 → MiniMax H3 → Qwen3-TTS → RTX upscale, all in batch (free workflow)
retained text
I've been working on a pipeline to make short history documentaries completely locally, and I wanted to share how it works. The workflow is in the description of the YouTube videotutorial: [https://www.youtube.com/watch?v=oOq5EsY6nTU](https://www.youtube.com/watch?v=oOq5EsY6nTU) 1. \*\*Script + prompts:\*\* a master prompt in Claude researches the topic (with sources), builds a "visual bible" (fixed style, character and location blocks so characters stay consistent) and writes the script in 10-second chunks. It outputs a zip with one prompt .txt per image (positive / negative / name), the video prompts and a Narration.txt. 2. \*\*Images:\*\* Load Prompts From Dir (Inspire Pack) feeds Qwen Image 2.1 in batch, one image per file. 3. \*\*Quality control:\*\* I upload the images back to Claude, it flags bad ones (hands, period errors, continuity) and rewrites only those prompts. Delete the bad image, set the index, regenerate. 4. \*\*Video:\*\* MiniMax H3 image-to-video, the approved image is the exact first frame, 10 s per clip, batch with an index + queue count. 5. \*\*Narration:\*\* Qwen3-TTS voice design → save the voice → batch TTS, one audio file per line. 6. \*\*2K upscale:\*\* RTX Video Super Resolution x2 on every clip. \*\*Hardware:\*\* RTX 5090 Laptop (24 GB VRAM, 64 GB RAM). Generation time of 10 minutes AI documentary: Images: 1 hours, Video Clips: 4 hours, Narration: 3 mimnutes, Edition: 15 minutes
document fetched 2026-10-27 00:42 · canonical https://reddit.com/r/comfyui/comments/1xfx5sv/automated_ai_2k_documentaries_in_comfyui_qwen