🎬 ABOUT THIS VIDEO
Tired of AI video drift, awkward cuts, and losing environment consistency between shots? In this tutorial, I’ll show you how to generate LONG, seamless MiniMax-H3 video continuations inside ComfyUI with 100% visual context consistency!
I built custom nodes specifically to solve the biggest problem in multi-shot AI filmmaking: quality degradation. Instead of re-encoding compressed MP4 files (which causes dark artifacts and generational loss), this workflow saves and chains the raw video latents directly—giving you seamless, continuous scenes without visual drift. In this video,
you'll learn:
✅ How to set up and prompt MiniMax-H3 using custom system prompts in Gemini
✅ The secret to chaining shots using saved latents (with zero quality loss)
✅ "Seed Video" vs. "Seed Ref Video" — how to maintain wide room layouts and subject positioning
✅ Pro-tips on subject weight balancing when introducing new characters
✅ Auto-trimming and stitching clips together into one continuous master take
✅ A quick sneak peek into the Version 2 one-shot pipeline I'm currently building! Both the custom nodes and complete workflow are 100% FREE to download
Description
FAQ
Comments (2)
Hi, setting up to use your Workflow.
I have some questions.
1) Where does: "h3_system_prompt_for_video_generation.md" come from?
edit: Oh, it's in some obscure location in a comment under the video linking to a patreon page with these prompts in the paid section.. o_O
2) Is this text-encoder in the workflow:
Minimax_H3_Qwen3vl_32b__nvfp4_awq.safetensors
The renamed version of:
https://huggingface.co/Comfy-Org/MiniMax-H3/blob/main/text_encoders/qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
yeah text encoder is same, also i just made my patreon and had nothing on it so i just put the system prompt there, not really hard to make your prompt, also i left the same prompt used in the workflow didnt remove it, so you can just edit it ,