🎬 MiniMax H3 + RelayVSR (Wan 2.2 5B) Cinema Super-Resolution Workflow

🌟 About this Workflow

A high-efficiency, production-ready video pipeline combining MiniMax H3 ultra-fast base generation with RelayVSR (powered by Wan 2.2 5B + Flash Decoder) native disk-to-disk super-resolution.

Special Credits & Big Thanks to Academia SD

Huge credit and sincere appreciation to Academia SD for the outstanding development of the native RelayVSR integration nodes for ComfyUI (comfyui_AcademiaSD)! His engineering work allows processing long videos in sequential chunks directly to disk with zero VRAM exhaustion, making high-end video super-resolution accessible to everyone. - 📺 Academia SD YouTube Channel: @Academia_SD - 🎥 Detailed Tutorial & Video Breakdown: RelayVSR Video Guide by Academia SD - 💻 Custom Node GitHub Repository: comfyui_AcademiaSD

✨ Key Features:

• Ultra-Fast Base Generation (Pass 1): Generates high-motion base video in just 8 turbo steps. Base resolution is customizable in the Resolution Standardizer (recommended: ~0.3 to 0.5 Megapixels, default ~0.4 MP).

• Synchronized Audio & Video: Decodes both video and audio latents simultaneously and automatically multiplexes them into a 24 FPS stream.

• RelayVSR Super-Resolution (Pass 2): Toggle between 2X (fast/preview) and 4X (full cinema HD) with a single click. Powered by Wan 2.2 5B keyframe conditioning + Flash Decoder interpolation, completely eliminating tiling seams and temporal flickering.

• VRAM Friendly: Sequential chunking keeps VRAM usage lightweight and stable even on 12GB - 16GB GPUs.

📦 Required Models & Download Links

1. RelayVSR Models (Wan 2.2 5B + Flash Decoder)

Place in ComfyUI/models/diffusion_models/: - 4X Model: relayvsr_x4_int8_convrot.safetensors (~4.81 GB) - 2X Model: relayvsr_x2_lightx2v_int8_convrot.safetensors (~4.81 GB)

Place in ComfyUI/models/vae/: - 4X Flash VAE: relayvsr_flash_vae_x4_fp32.safetensors (~128 MB) - 2X Flash VAE: relayvsr_flash_vae_x2_fp32.safetensors (~128 MB)

2. MiniMax H3 Base Models

Place in ComfyUI/models/diffusion_models/: - H3 Model: MiniMax_H3_FL2VA_pruned_nvfp4.safetensors (or any MiniMax H3 NVFP4/GGUF checkpoint) Place in ComfyUI/models/clip/: - Text Encoder: qwen3vl_32b_heretic_minimax_h3_nvfp4.safetensors Place in ComfyUI/models/vae/: - Video VAE: minimax_h3_video_vae_fp16.safetensors - Audio VAE: minimax_h3_audio_vae_fp32.safetensors

3. Required Custom Nodes

Install via ComfyUI Manager: - comfyui_AcademiaSD by AcademiaSD - ComfyUI-VideoHelperSuite by Kosinkadink - ComfyUI-Custom-Scripts by pythongosssss

🚀 How to Run:

1. Load your initial image into 1. Start Frame.

2. Enter your creative prompt in the H3 Promptor (or use natural language; the built-in translator handles the rest).

3. Set your target base resolution (default 0.4 MP) in Resolution Standardizer.

4. Choose 2X or 4X on the RelayVSR Resolution Switch at the top.

5. Click Queue Prompt and enjoy cinema-grade video with original audio!