Instructions to use WarmBloodAban/Minimax_H3_LoRAs with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use WarmBloodAban/Minimax_H3_LoRAs with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline from diffusers.utils import load_image, export_to_video # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("MiniMaxAI/MiniMax-H3", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("WarmBloodAban/Minimax_H3_LoRAs") prompt = "-" input_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png") image = pipe(image=input_image, prompt=prompt).frames[0] export_to_video(output, "output.mp4") - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
Minimax-H3 LoRAs

- Prompt
- -
Model description
license: apache-2.0 language: - en - zh pipeline_tag: image-to-video tags: - minimax-h3 - lora - video-generation - text-to-video - image-to-video - comfyui - game-cg - third-person-view
MiniMax-H3 Concepts LoRA Series
Welcome to the MiniMax-H3 Concepts LoRA Series repository! This collection features fine-tuned Low-Rank Adaptation (LoRA) models built upon the MiniMax-H3 video generation framework. This repository is dedicated to exploring specialized visual concepts, camera controls, and gaming aesthetic enhancements, and will be updated continuously.
๐ข Release Notes & Updates
- [2026/09] Initial release of `Minimax-h3_Third_person_view` โ a specialized LoRA for game CGs, HUD UIs, and dynamic 3rd/1st-person camera controls.
๐ฆ Model Catalog
1. Minimax-h3_Third_person_view
๐ Overview
`Minimax-h3_Third_person_view` is specifically fine-tuned for cinematic game cutscenes, HUD interfaces, and dynamic third-person / first-person perspective transitions.
- Base Model: Fine-tuned on top of the MiniMax-H3 INT8 Reference architecture.
- Core Features:
- Precise control over Third-person view and Over-the-shoulder camera tracking (spring-arm camera motion, over-the-back angles, and dynamic distance adjustments).
- Seamless support for First-person perspective (First-person POV / FPV) and smooth camera whip-pans or view transitions.
- Native generation of game HUD overlays, target-lock UI reticles, QTE prompts, Boss health bars, and high-impact action VFX.
๐ฌ Video Demonstration
๐ Prompt Guide & Examples
For optimal performance, structured prompt formatting (such as full-reference or structured shot breakdowns) is recommended when using ComfyUI or API calls:
subject_definitions:
<Subject 1> is the cute mint-green and pastel-orange cat mecha character from <Picture 1>, featuring a rounded pearlescent feline head with large blue glowing optic eyes, metallic whiskers, floral/star decals, a paw-print chest emblem, exposed dark mechanical joints, and a flowing tattered beige cape.
<Subject 2> is the pale, eyeless wyvern-like monster from <Picture 3>, featuring a long muscular neck, a circular ring-like gaping maw lined with multiple rows of sharp teeth, smooth rubbery pale skin with blueish undertones, clawed wings, and heavy bipedal limbs.
<Environment 1> is the eerie ruined village setting from <Picture 2>, featuring a wooden entrance gate with lanterns, blood-red spider lilies, red vines, wet stone paving, and thick fog.
<Picture 1> is the character design reference sheet for <Subject 1>.
<Picture 2> is the environment design reference sheet for <Environment 1>.
<Picture 3> is the monster design reference sheet for <Subject 2>.
summary:
[reference generation] A 10-second photorealistic third-person action RPG Boss encounter gameplay sequence in <Environment 1>, where <Subject 1> dodges a leap attack from <Subject 2>, counterattacks with an energy blade combo, parries a wing-claw swipe, and charges an ultimate finisher, tracked continuously by an over-the-shoulder camera.
retention_analysis:
<Subject 1> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4], [Shot 5], [Shot 6]): fully_preserved - the mint-green and pastel-orange cat mecha chassis, blue glowing optics, paw emblem, and tattered beige cape from <Picture 1> are fully preserved.
<Subject 2> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4], [Shot 5], [Shot 6]): fully_preserved - the pale rubbery skin, circular tooth-filled gaping maw, long neck, and winged beast design from <Picture 3> are fully preserved.
<Environment 1> (appears in [Shot 1], [Shot 2], [Shot 3], [Shot 4], [Shot 5], [Shot 6]): fully_preserved - the ruined village architecture, wooden entrance gate, red spider lilies, wet cobblestones, and foggy atmosphere from <Picture 2> are fully preserved.
detailed_description:
The target video features a photorealistic 3D third-person action RPG Boss battle gameplay aesthetic rendered with Unreal Engine game mechanics, deep depth of field, over-the-shoulder spring-arm camera tracking, and transparent combat UI elements in the top-right corner.
[Shot 1] The video opens with an over-the-shoulder camera locked 3.0 meters directly behind <Subject 1> in <Environment 1>. <Subject 2> screeches with its gaping circular maw wide open and leaps off a ruined rooftop above the wooden gate, descending with extended clawed wings toward <Subject 1>. The camera tilts slightly upward to track the falling trajectory of <Subject 2>.
[Shot 2] At 00:01.500, back thrusters on <Subject 1> burst with blue glowing energy, driving a rapid rightward side-dodge. <Subject 2> slams its heavy pale body into the wet stone paving, shattering cobblestones into flying debris and triggering a heavy camera impact shockwave shake while maintaining over-the-shoulder tracking.
[Shot 3] At 00:03.000, as <Subject 2> pauses in a brief recovery animation with its long neck writhing, <Subject 1> plants its mechanical boots, draws a glowing mint-green high-frequency energy sword, and whips its beige cape back. The camera pushes forward 15% closer behind <Subject 1>'s back.
[Shot 4] At 00:04.500, <Subject 1> executes a fast three-hit energy combo slash into <Subject 2>'s pale rubbery torso and neck, releasing brilliant sparks and energy particle flashes while <Subject 2> flinches back in pain. Subtle hit-impulse camera micro-shakes accompany each slash.
[Shot 5] At 00:06.500, <Subject 2> counters with a violent swipe of its left wing-claw, but <Subject 1> instantly parries with its energy blade. The weapon collision erupts into a radiant light shockwave that blows red spider lily petals into the air across <Environment 1>.
[Shot 6] At 00:08.500, <Subject 2> reels back in a heavy stagger, opening its tooth-lined mouth wide. <Subject 1> steps back into a charging stance, overloading its sword with pulsing energy as a golden finisher UI targeting icon snaps onto <Subject 2>, ending the shot at 00:10.000.
overall_soundscape:
Monstrous wet roars, heavy ground impact thuds, thruster ignition bursts, metallic blade unsheathing, energy slash impacts, parry shockwave clashes, scattered foliage wind rustles, and crisp action RPG combat UI audio effects.
non_diegetic_music:
An intense, high-tempo epic Boss battle orchestral-electronic track featuring heavy industrial percussion and low pulsing synth bass that swells in intensity as the parry and finisher charge.
๐ก Key Trigger Words & Recommended Tags
Camera Views: third-person perspective, over-the-shoulder camera, first-person POV, FPV HUD
Game Rendering: rendered in Unreal Engine, gameplay sequence, combat stance, macro close-up
UI Elements: transparent HUD elements, target-lock UI reticle, QTE UI prompt, floating damage text UI
โ๏ธ Recommended Inference Settings
Base Model: MiniMax-H3 INT8 Reference Model
LoRA Weight: 0.6 โ 0.85 (0.7 is recommended as a starting point)
Compatible Workflows: ComfyUI MiniMax-H3 native nodes and official API endpoints.
๐ฎ Future Roadmap
This repository will be actively maintained. Upcoming updates include:
[ ] Additional stylized environment & cinematic LoRAs
[ ] Camera motion & action dynamics enhancement adapters
If you encounter any issues or have questions, feel free to open a thread under Community / Discussions!
๐ค Community & Commercial Inquiries
Feel free to connect for tutorials, community discussions, workflow sharing, or commercial collaborations:
- YouTube Channel: AIGC-Singularity
- Bilibili Channel: AIGC-Singularity Space
- QQ Group 1: `1058747239` (Request to join)
- QQ Group 2: `1072010342` (Request to join)
- Business Inquiries (WeChat): `aigctyd`
- Email: `a592991299@gmail.com`
Download model
Download them in the Files & versions tab.
- Downloads last month
- 242
Model tree for WarmBloodAban/Minimax_H3_LoRAs
Base model
MiniMaxAI/MiniMax-H3