Introduction

If your 8GB card has been side-eyeing MiniMax H3 like “maybe someday,” this one’s for you. ArtRealmAI is dropping a ready-to-queue ComfyUI workflow: MiniMax H3 — 10Eros + Semantic Bridge + Ultimate Upscale — a two-stage text-to-video graph that cooks a short native-audio clip, then lifts it with the official H3 3D latent upscaler.

Download the workflow JSON, drop the files, restart ComfyUI, queue. You get the original clip and the upscaled clip.

Workflow download:

minimax-h3-10eros-semantic-bridge-ultimate-upscale.jsonJSON · 124.32 KB

What this workflow is

Three ideas, one graph:

  1. Stage 1 — 10Eros TURBO-hybrid generates a short AV clip at 0.3–0.4 MP in 6 steps (euler / simple). Turbo is already baked into 10Eros — do not stack LightX2V, ModelTC, or other speed LoRAs on top.
  2. Semantic Bridge — an ~11 MB conditioning adapter (not a second video model) that blends a learned semantic residual into H3 text conditioning. Think tighter prompt structure: spatial relations, stated actions, materials. Not a style LoRA. Not a VRAM tax.
  3. Stage 2 — MMH3 Ultimate Upscale with the official H3 3D latent upscaler, 8GB-friendly tiles, denoise 0.18.

H3 is CFG-distilled. This graph uses BasicGuider — no negative prompt, no CFG. Put exclusions in the positive prompt.

SenseNova is not required at runtime. Semantic Bridge was distilled from a teacher; inference is H3 + the tiny adapter only.

Custom nodes (just two extra packs)

Everything else is ComfyUI 0.33+ native MiniMax H3.

  1. Comfyui-MMH3-UltimateUpscale → custom_nodes/
  2. MiniMax H3 Semantic Bridge — extract MiniMax_H3_Semantic_Bridge_v1.0.zip into custom_nodes/ (not in ComfyUI-Manager)

Restart ComfyUI after both are in place.

Files to download

diffusion_models10Eros_Max_h3_TURBO-hybrid_beta3_int8_convrot_skip_edges.safetensors

text_encodersqwen3vl_32b_minimax_h3_nvfp4_awq.safetensors

vaeminimax_h3_video_vae_fp16.safetensorsminimax_h3_audio_vae_fp32.safetensors

latent_upscale_modelsminimax_h3_latent_upscaler_3d_fp16.safetensors

semantic_bridge (models/semantic_bridge/)MiniMaxH3_SemanticBridge_v1.safetensors

CLIPLoader type = minimax.

ComfyUI/├── custom_nodes/│ ├── MiniMax_H3_Semantic_Bridge/│ └── Comfyui-MMH3-UltimateUpscale/└── models/ ├── diffusion_models/ 10Eros skip-edges INT8 ├── text_encoders/ qwen3vl NVFP4 ├── vae/ video fp16 + audio fp32 ├── latent_upscale_models/ H3 3D upscaler fp16 └── semantic_bridge/ MiniMaxH3_SemanticBridge_v1.safetensors

8GB defaults (as shipped)

Setting              Stage 1                          Stage 2
-------------------  -------------------------------  --------------------------------------
Size                 9:16 @ 0.4 MP                    9:16 @ 0.9 MP (1280×736)
Sampler              euler / simple / 6               euler / simple / 6
Denoise              1.0                              0.18
Attention            Comfy Kitchen                    same model
Semantic Bridge      alpha 0.10, per_token            same prompt, matching W/H
Tiles                —                                512×384 (set ≥ upscale size to disable)
Temporal             —                                chunk 1020 = off for a 5s clip

Launch tip: --lowvram --force-fp16 --use-ck-attention

Load Image is bypassed = text-to-video. Enable it and wire first_frame only when you want I2V.

Outputs land at video/MiniMax_H3_Original and video/MiniMax_H3_UltimateUpscale.

How to use it (quick path)

  1. Download the workflow JSON with the button below
  2. Install the two custom node packs above and restart ComfyUI
  3. Drop the listed weights into the folders shown
  4. Load / drag the JSON into ComfyUI
  5. Keep the bundled golden-hour causeway prompt (or swap your own structured H3 prompt)
  6. Queue — grab both the original and upscaled saves
minimax-h3-10eros-semantic-bridge-ultimate-upscale.jsonJSON · 124.32 KB

Prompt format

H3 wants a structured prompt, not a tag salad. The bundled example is a single continuous golden-hour shot on a wooden beach causeway (sundress, ocean, no on-screen text), with overall_soundscape and non_diegetic_music so the native audio track has somewhere to go.

Keep Semantic Bridge alpha at 0.10 first. 0.15 is stronger (what the adapter author used for A/B clips) — it can help stated constraints, and it can over-steer. It’s experimental on 10Eros (distilled on native H3 FL2VA conditioning, which this graph still uses).

Do not use this adapter on Ref2VA.

Pro Tip

Resist the urge to “help” 10Eros with extra turbo LoRAs or a negative prompt. Turbo is already inside the checkpoint, and H3’s distilled guider ignores negatives while still charging you a second pass. If VRAM dies on a longer clip, drop temporal chunk_length to 85 with overlap 17 before you start gutting quality. Upscale denoise above 0.18 tends to get weird fast.

Tips worth tattooing on the node graph

  • No turbo / speed LoRAs on 10Eros
  • No negative prompt
  • Extra style LoRAs: use ComfyUI’s built-in Load LoRA (Model Only) after UNET — this share doesn’t ship unpublished optional-LoRA nodes
  • Higher upscale denoise than 0.18 → artifact city

Credits

  • MiniMax H3 — MiniMax / Comfy-Org packages
  • 10Eros TURBO-hybrid — TenStrip / cicalooo INT8 skip-edges
  • Semantic Bridge adapter + nodes — speach1sdef178
  • MMH3 Ultimate Upscale — bbaudio-2025
  • Graph — ArtRealmAI / artrealmai.com

License (read this)

Free to use for personal and commercial generation. Videos you create with this graph are yours.

You may not sell, rebundle, paywall, or republish this workflow JSON (or a trivial copy) without permission from ArtRealmAI. Ask at artrealmai.com.

Upstream MiniMax H3 Community License still applies to the H3 weights and to Semantic Bridge as an H3 model derivative. 10Eros and the upscaler have their own pages.

Conclusion

Consumer VRAM, native audio, Semantic Bridge for structure, Ultimate Upscale for the finish — that’s the ArtRealmAI H3 share in one sentence. Grab the JSON, load the stack, and go make something golden-hour gorgeous.

Download the workflow:

minimax-h3-10eros-semantic-bridge-ultimate-upscale.jsonJSON · 124.32 KB

—Aurelia ♡