keys-2k MiniMax-H3 — Parallel processing for Two DGX Sparks
One-command pull · pinned quality stack · master-K0 dual-Spark parallel · ESRGAN “2K” · Spectrum audio fix · Contex-Loop / MultiRef · optional realism LoRA
Dual-serve DS4 + dual H3 on two DGX Sparks was made possible by Tony (tonyd2wild) .keyspark quality specialization (ablit co-tenant profile + sealed bf16 finals with stock int8-convrot TE — heretic retired — + farm-mode dual parallel + 2K workflows).
Why this exists
Agents and users often git clone old Spectrum / Motion Context / Comfy builds and lose:
Spectrum v0.2.1 audio fix (offline_smoothing_replay=true)
Contex-Loop / MultiRef seam fixes
Legal native sizes for upscale-to-2K
Dual-node master-K0 parallel spans
This package pins the full integrated stack in a Docker image + manifest so the default path is the upgraded one.
One-command pull
curl -fsSL https://huggingface.co/drowzeys/keys-2k-MiniMax-H3-Parallel-Two-DGX-Sparks/raw/main/scripts/one_command_pull.sh | bashOr:
1 git clone https://huggingface.co/drowzeys/keys-2k-MiniMax-H3-Parallel-Two-DGX-Sparks
2 cd keys-2k-MiniMax-H3-Parallel-Two-DGX-Sparks
3 bash scripts/one_command_pull.sh
Pull Docker only:
1 docker pull ghcr.io/drowzeys/keys-2k-minimax-h3-parallel-two-dgx-sparks:0.31.1-pp20260811
2 # also tagged :latest
What’s in the image (pinned)
Component Version / pin ComfyUI 0.31.1 Spectrum MiniMax H3 0.2.1 (audio fix defaults)Contex-Loop 0.3.8 Motion Context + MultiRef pinned SHAs in manifest/H3_STACK_MANIFEST.json Sol-attn Blackwell + Triton SolAttn pinned KJNodes (Sage) 1.4.9 · sageattention==1.0.6 NKD Preview Tools 3.3.0 Launch flags --disable-pinned-memory --reserve-vram 48 --vram-headroom 10
Not for quality: Turbo / few-step LoRA graphs (present optional; leave off).
See manifest/H3_STACK_MANIFEST.jsonscripts/verify_h3_stack.sh
2K path (open weights)
Open MiniMax-H3 does not native 2K in one pass. This package’s production path:
1 Native denoise 704 × 1280 (multiple of 32 — 720×1280 fails)
2 ↓ RealESRGAN ×2 (inline on spans)
3 Delivery 2K ~1408 × 2560
Parallelism: master-K0 keyframes + parallel FL2VA spans on two Sparksh3-spans.py --kf-mode master-parallel --upscale).
Workflow package: workflows/anime_2k_bench/0811_012837: ~18.8 min wall for ~4.8 s @ 704×1280 dual-node (co-tenant DS4).
Dual DGX Spark — parallel H3
1 export HEAD = 10.100 .10.1 WORKER = 10.100 .10.5 # your fabric IPs
2 export MODELS_DIR = $HOME /keys-2k-minimax-h3-parallel/models
3
4 # weights once
5 bash scripts/fetch_weights.sh
6 # or: rsync -a /path/to/lab/h3-cotenancy/ComfyUI/models/ $MODELS_DIR/
7
8 # H3 on both nodes
9 bash scripts/run_dual_h3.sh
10
11 # quality 2K parallel bench
12 bash scripts/run_anime_2k_bench.sh
13 # optional realism LoRA (file in models/loras/):
14 REALISM = 1 bash scripts/run_anime_2k_bench.sh
Single node smoke:
1 docker run --rm --gpus all --network host \
2 -v $MODELS_DIR :/opt/powerpack-h3/ComfyUI/models \
3 -v $HOME /Videos/h3-out:/opt/powerpack-h3/ComfyUI/output \
4 ghcr.io/drowzeys/keys-2k-minimax-h3-parallel-two-dgx-sparks:0.31.1-pp20260811
5 # UI/API: http://<host>:8188
Compose: compose/docker-compose.yml
Co-tenancy with DeepSeek-V4-Flash (Tony’s rule)
If running DS4 + H3 on the same two Sparks:
DS4 first until /v1/models healthyH3 second (this image)Power Pack util 0.76 / 888k (or ≤ 0.85 hard cap)
Teardown: H3 → DS4
One heavy video job per Spark (concurrency=2 = two boxes)
Full dual-boot recipe: Power Pack deploy/keyspark/bringup.sh.
A/V glitch mitigations (baked into graphs)
Fix Setting Spectrum speech stutter offline_smoothing_replay=true, audio_blend_weight=0Multishot seams Motion Context / Contex-Loop / MultiRef Long-gen hallucination master-K0 short parallel spans (not one long gen)
Weights (not fully redistributed here)
Place under models/ (bind-mount into the container):
File Role diffusion_models/minimax_h3_fl2va_pruned_bf16.safetensorsFL2VA DiT — FINALS (sealed) diffusion_models/minimax_h3_ref2va_pruned_bf16.safetensorsref2va — FINALS (sealed) diffusion_models/minimax_h3_fl2va_pruned_int8_convrot.safetensorsFL2VA DiT — draft tier only diffusion_models/minimax_h3_ref2va_pruned_int8_convrot.safetensorsref2va — draft tier only text_encoders/qwen3vl_32b_minimax_h3_int8_convrot.safetensorsSTOCK int8-convrot TE (sealed; heretic retired)vae/minimax_h3_video_vae_fp16.safetensorsvideo VAE vae/minimax_h3_audio_vae_fp32.safetensorsaudio VAE upscale_models/RealESRGAN_x2plus.pth2K upscale loras/h3-realism-people-t2v-i2v-r2v.safetensorsoptional realism
Helpers:
scripts/fetch_weights.sh
Upstream:
Comfy-Org/MiniMax-H3 ,
Kijai/MiniMax-H3_comfy .
Verify (agents must pass)
1 bash scripts/verify_h3_stack.sh 10.100 .10.1
2 bash scripts/verify_h3_stack.sh 10.100 .10.5
Fails if Spectrum audio-fix default is wrong or core nodes are missing.
Credits
Dual-serve co-tenancy Tony / tonyd2wild — ds4-h3-video-gen-factory Spectrum v0.2.1 audio fix xmarre / ComfyUI-Spectrum-MiniMax-H3 Motion Context NikoDemon80 MultiRef seitanism Contex-Loop ethanfel NKD Preview Nekodificador Sol / Sage / heretic NVIDIA Sol-engine community · KJNodes · sakamakismile et al. Packaging keyspark / @drowzeys Power Pack
License
Package scripts/docs: see Power Pack repo
Model weights: their respective licenses (not all redistributed here)
ComfyUI and custom nodes: upstream licenses (GPL where applicable)
