MiniMax H3 RefMod Two-View Scene

https://youtu.be/H5L5sO7dRu8

Creator: AI-KSK · Collection: Video 327 / six-workflow MiniMax H3 RefMod episode

Workflow file: 04_武道馆场景RefMod_双视角批量.json · Cover file: 暖光武道馆双视角参考-4.png

Main features:

This workflow makes a reusable scene reference from two images of the same location. The example location is a warm, wood-built martial arts dojo, with lanterns, lattice windows, architectural beams, weapon racks and a ceremonial drum. The aim is to preserve a recognizable set across multiple shots while changing action and performers—not to reconstruct a physically exact 3D room.

Unlike the motion graph, the two-reference configuration really is connected: two LoadImage nodes feed refs_image.ref_image_0 and refs_image.ref_image_1 of MiniMaxH3RefModExtract, then MiniMaxH3RefModInspect provides preview. The two serialized JPEG filenames start Gemini_Generated_Image_...; neither original JPEG is included here.

Exact values from this JSON

- name=Dojo_OneLocation, mode=Full Reference, concept_type=background.
- ref_resolution=1024, pool_h=16, pool_w=16, latent_frames=16.
- motion_only=false, save=true, subfolder=scene.
- The text description requests one empty dojo, persistent architectural details and consistent warm light.

Suggested workflow:

Supply your own two scene views. Use photographs or renders you have permission to use; ensure both represent one recognizable room. Update both LoadImage selections and, if necessary, rewrite the description so it accurately matches the new environment. Run the graph, inspect the created scene/Dojo_OneLocation reference and load it in the generation workflows. The scene cover shown with this project is not a substitute for the two missing JPEG inputs.

Limits

Two-view reference conditioning can encourage visual continuity, but does not provide camera-calibrated geometry, guaranteed room consistency or a digital twin. The golden dojo and split-angle panels on the cover are promotional design.

Dependencies, tutorial and deployment

This project uses ComfyUI, MiniMax H3 weights and the upstream RefMod node pack; workflows involving VHS_ nodes require Video Helper Suite. See Comfy-Org's H3 model inventory for official distribution guidance. The exact int8-convrot video VAE and Turbo 4-step LoRA named in the creator's JSON need separate availability/provenance checks. We distribute this workflow JSON and its original cover, not external checkpoints or third-party media. Respect applicable rights and licenses.

Deploy AI-KSK's H3｜RefMod免训练超级参考 image on 优云智算. Referral disclosure: the deployment link includes an AI-KSK referral code and may support the creator. Check the cloud provider's actual billing and environment before deploying.

Watch the complete six-workflow YouTube tutorial. This project is one module in that episode. Description and settings were prepared by inspecting the source JSON and transcript, not from a new measured benchmark.

Bilibili tutorial: https://www.bilibili.com/video/BV1F2nZ6SEA7/