Z-Image × H3 PRO

Z-Image Turbo paints the still. That still becomes the MiniMax H3 first frame. H3 adds motion and native stereo. One queue. I2VA by design.

Z-Image runs at 1 MP, 8 steps, CFG 1. H3 runs at 0.2 MP, then a 2× latent refine and FILM to 48 fps.

Two LoRA stacks: Z-Image on ND Super LoRA Loader, H3 on LTX LoRA Loader (mode normal). Empty stacks are fine.

Custom nodes

KJNodes — DiffusionModelLoaderKJ, MiniMaxH3ChunkFeedForward, MiniMaxH3SigmaShift, ImageSharpenKJ, ModelPreviewOverrideKJ
https://github.com/kijai/ComfyUI-KJNodes

PlagueKind Nodes — LTX_lora_loader
https://github.com/PlagueKind/ComfyUI-PlagueKind-Nodes

ND Super Nodes — NdSuperLoraLoader
https://github.com/HenkDz/nd-super-nodes

ClipProj — ClipProjLoader
https://github.com/nicolab28/ComfyUI-ClipProj

SolAttn — SolAttnPatch
https://github.com/sumeetprashant/ComfyUI-SolAttn

Minimax H3 Latent Upscaler 3D — MinimaxH3LatentUpscaler3D
https://github.com/xmarre/Comfyui_Minimax_h3_latent_Upscaler-Plus

H3 latent / cond upscale — MiniMaxH3ConditioningUpscale
https://github.com/rockerBOO/h3-latent-upscaler

ComfyUI-GGUF — CLIPLoaderGGUF
https://github.com/city96/ComfyUI-GGUF

Memory Cleanup — RAMCleanup, VRAMCleanup
https://github.com/LAOGOU-666/Comfyui-Memory_Cleanup

NVIDIA RTX Nodes — RTXVideoSuperResolution
https://github.com/Comfy-Org/Nvidia_RTX_Nodes_ComfyUI

Needs ComfyUI 0.30+ for native MiniMax H3.

Models

Z-Image
zimageTurboByStable_2602BF16.safetensors
qwen/Qwen3-4B.fp16.gguf
flux1_vae_bf16.safetensors

H3
minimaxh3/10Eros_Max_h3_TURBO-hybrid_beta5_int8.safetensors
minimax_h3_video_vae_int8_convrot.safetensors
minimax_h3_audio_vae_fp32.safetensors
qwen3-vl-4b-heretic_nvfp4.safetensors
mmh3-4b-ClipProj-celeb-mlp.safetensors
minimax_h3_latent_upscaler_3d_bf16.safetensors
film_net_fp16.safetensors
taeh3.safetensors

Put FILM in models/frame_interpolation. Put ClipProj matrices in models/clip_projections. Put the latent upscaler in models/latent_upscale_models.

if you have any questions please let me know!