MiniMax H3 Latent Upscaler | Two-Pass Video Generation

Generate higher-resolution videos from text or reference images with built-in upscaling and second-pass refinement.

Who it's for: creators who want this pipeline in ComfyUI without assembling nodes from scratch. Not for: one-click results with zero tuning - you still choose inputs, prompts, and settings.

Open preloaded workflow on RunComfy

Open preloaded workflow on RunComfy (browser)

Why RunComfy first
- Fewer missing-node surprises - run the graph in a managed environment before you mirror it locally.
- Quick GPU tryout - useful if your local VRAM or install time is the bottleneck.
- Matches the published JSON - the zip follows the same runnable workflow you can open on RunComfy.

When downloading for local ComfyUI makes sense - you want full control over models on disk, batch scripting, or offline runs.

How to use (local ComfyUI)
1. Load inputs (images/video/audio) in the marked loader nodes.
2. Set prompts, resolution, and seeds; start with a short test run.
3. Export from the Save / Write nodes shown in the graph.

Expectations - First run may pull large weights; cloud runs may require a free RunComfy account.

Overview

MiniMax H3 Latent Upscaler generates higher-resolution videos with audio from text prompts and optional reference images. This BUNNYH3 two-pass ComfyUI workflow first generates the scene and motion at a lower resolution, then upscales the video's latent representation and runs a second sampling pass to refine the enlarged result before decoding. Use it to create short product ads and cinematic scenes with adjustable output resolution. Upscaling is built into video generation; this is not a general-purpose tool for restoring uploaded videos.

Important nodes:

Key nodes in Comfyui MiniMax H3 Latent Upscaler workflow

• Stage 1 Conditioning · AUTO (#7). Central entry for text, references, and mode routing. Adjust prompt and, when needed, the mode that governs how references are used. Keep this on AUTO unless you have a clear reason to force a mode, and ensure references match the story you describe.

• Stage 1 Dual Clock · Video 12 / Audio 3 (#8). Orchestrates separate sampling clocks for video and audio to keep motion beats and sonic events aligned. If you must retime, do so gently and keep both clocks consistent with your target pacing.

• H3 3D Learned Latent Upscale (#13). Applies the MiniMax H3 Latent Upscaler using the official 3D fp16 checkpoint. The only parameter most users should touch is the scale factor; increasing it raises width and height in latent space, and total pixels rise with the square of that factor. Use the upscaler’s validated checkpoint from the model card to avoid corrupted latents. Hugging Face

• Stage 2 Reconcile · Size & Audio (#15). Ensures the Stage 2 latent carries the upscaled size and the selected audio policy before reconstruction. If you experiment with audio settings, keep reconciliation enabled so timing and dimensions remain consistent.

• Stage 2 Detail Mixer · Fixed 5 Steps (#16). The heart of high-resolution reconstruction that adds detail without rewriting motion logic. If the result looks oversharpened or drifts in animation, reduce Stage 2 LoRA strengths first, then fine-tune mixer balances only if necessary.

• Stage 2 AV Decode (#20). Decodes the reconstructed latent into frames and synchronized audio. Always route outputs from this node into CreateVideo and Save Final Video to ensure color space and audio are handled correctly.

Notes

MiniMax H3 Latent Upscaler | Two-Pass Video Generation - see RunComfy page for the latest node requirements.