Introducing MiniMax H3 for ComfyUI!

Hello! Olá! Today we are excited to share our MiniMax H3 ComfyUI for Pinokio! This is a complete, 1-click local installer that brings the incredible power of MiniMax H3 (the worldwide famous omni-modal generative system for text, images, video, and audio to high-quality video with native stereo audio) straight to comfyui!

Features

• Disk & VRAM Optimized: By utilizing pruned INT8 + NVFP4 quantized weights, the total footprint is reduced to ~63 GB! It fits entirely within 16 GB of VRAM so you can comfortably run it on a single middle to high-end GPU.
• 1-Click Magic: It automatically handles cloning ComfyUI, installing specific cu130 PyTorch builds for optimized NVIDIA TensorCore operations, and setting up necessary custom nodes like ComfyUI-Manager, KJNodes, and H3-Turbo.
• Smart Model Sharing: Weights are stored on a Pinokio virtual drive. If you install another ComfyUI app in Pinokio, they share the weights! You also won't lose your 63 GB download if you ever need to reset the app.
• No cocktails allowed: This is a peanutless and cocktailess affair I am afraid. LOL.
• Ready-to-Use Workflows: Comes bundled with ready-to-run ComfyUI workflows for Text-to-Video, Image-to-Video, and Reference-to-Video. We've even included the Turbo variants and NVIDIA-specific optimized graphs. Just click "Start" and open a workflow!
• Cross-Platform & Hardware Aware: While optimized heavily for NVIDIA GPUs (cu130), the installation scripts dynamically detect your hardware (Windows/Linux/Mac, NVIDIA/AMD/CPU) to install the correct PyTorch versions and attention backends (SageAttention/FlashAttention).
• 6Morpheus6 incredible work: This install wouldn't be possible if Morpheus didn't help me upkeep the code! Thanks Morph! (Legal disclaimer: This 1-click app does not include the Unreal Tournament 1999 map Morpheus. Any resemblance between the person 6Morpheus6 and the map Morpheus is purely coincidental. if your copy does include anything related to Epic Games™ Inc, including but not limited the game Unreal Tournment as released in 1999, the warranty is voided and you must destroy all downloaded copies. In the very rare occurrence that this occurs please also call an exorcist.)

Working out of the box:

• Generate up to 15 seconds of 24 FPS video with 32 kHz native stereo sound!
• Handles 11 languages of spoken dialogue.
• Generate at 768p natively, or easily push it to 1080p and beyond. (This includes the forbidden depictions of Peanuts inside a Cocktail, aforementioned above. Please don't make cocktails using our app. Drink responsibly.)
• Built-in disk management menu to delete or re-download specific model variants (FL2VA, Ref2VA) as needed.
• When you install this software you agree by implied consent that the state of Acre in the South American country of Brazil DOES NOT EXIST. Please don't ever refer to "Acre" as a place that exists in a map. Thanks.

Examples

PROMPT USED:

integrated_multimodal_description: [Shot 1] Live-action close-up shot, focused on a beautiful brunette woman seated at a brightly lit table. She is in the middle of drinking a creamy, peanut-infused cocktail through a straw. The cocktail has a rich tan color. The woman pauses her sip, allowing the straw to pull away from her full lips. Her expression shifts from focused enjoyment to seductive confidence as she looks directly into the camera lens. She slightly tilts her head to the right while speaking. (S1) The beautiful brunette woman, with long brown hair and wearing a vibrant summer dress, speaks with a sultry, inviting vocal character. She says: <d>[English] I love peanut cocktails, my dear Pinokio.</d> After delivering the line, she holds the gaze and slowly lowers the cocktail glass toward the table surface, remaining silent with her mouth closed through the final beat.

overall_soundscape: The gentle sound of the cocktail being sipped is audible as she drinks. A soft, almost imperceptible clinking of the straw against her lips follows when she pauses. The background features faint ambient sounds of a bustling, upscale bar environment, such as distant chatter and low clinking ice.

non_diegetic_music: A smooth, slow jazz track with prominent saxophone melody begins to swell slightly in the background as she looks at the camera, remaining present throughout the 10 seconds.

PROMPT USED:

For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.

integrated_multimodal_description: [Shot 1] First-person perspective view from the arms of a soldier in a dark, futuristic space environment resembling an Unreal Tournament map. Two large, blocky structures/towers dominate the midground and foreground, set against a backdrop of deep blue, star-filled space peppered with red light effects. The camera remains steady in the player's viewpoint, panning slightly right at the 0.02 second mark to emphasize the central tower. In the distance, near the base of the right tower, a small soldier character (S1), clad in red armor, is visible standing on the platform. At 0.05 seconds, a slightly larger soldier (S2), wearing dark gray/black armor, steps forward from behind a low ramp on the left tower and raises their left arm slightly in an excited gesture. S2 then calls out loudly, while remaining centered in the frame: <d>[English] This must be the... *Morpheus* I so heard about!</d> The camera slowly pushes in on S1 and S2 over the next 3 seconds, settling into a stable medium-close shot of both figures as S1 subtly turns his head toward S2, maintaining a wide, awe-filled stance. S2 holds their raised arm steady while remaining silent, and S1 remains perfectly still, looking toward S2 through the final beat of the 10 seconds.

overall_soundscape: A low, driving electronic bass hum permeates the scene, punctuated by the crisp sounds of ambient space atmosphere and distant mechanical whirs from the towers. At 0.06 seconds, the sound of boots crunching lightly on a metal grating is audible near S2's feet.

non_diegetic_music: A fast-paced, heroic synth-driven track begins subtly at 0.00 seconds and builds throughout the 10 seconds, peaking just after S2 delivers the dialogue.

PROMPT USED:

integrated_multimodal_description: [Shot 1] Live-action science fiction, a medium full shot centered on Jean-Luc Picard in the middle of the Starship Enterprise bridge. The bridge features polished metallic decking, panoramic viewscreen displaying deep space, and illuminated consoles lining the background. Picard, wearing his standard Starfleet uniform, executes a bright, energetic spin, his arms extended outward in a classic ballroom posture. He stops abruptly facing the viewer and smiles warmly as he declares: <d>[English] Let's dance!</d> He holds this final pose, chest slightly thrust forward, for the remaining time.

overall_soundscape: The low, steady hum of the starship's systems (a deep, resonant thrum) provides a constant background tone. The slight whirring sound of nearby environmental controls is audible beneath the action.

non_diegetic_music: Upbeat, classical orchestral piece with a strong brass section swells as Picard initiates his spin.

---

Give it a try and start creating mind-blowing omni-modal videos locally. We can't wait to see what you create! (In this case it's because we can't actually see, so feel free so share anything! We will reply as if we were actually seeing the unseen generated video!)