Model
Trends
.ai
Model
Trends
.ai
die besten Open-Source-KI-Modelle

MiniMax H3 Combat BASE V2: fight motion, impact and drama

Thema: Modelle
Von Captain
Veröffentlicht am 2026-10-01
FourBunny's Combat BASE LoRA for MiniMax H3 improves fight choreography, hit reactions, continuity and physical cause and effect, and it helps dialogue scenes too. Learn the prfight1, prfin1 and prslow1 triggers and how to stage a scene.
Diese Seite ist noch nicht übersetzt und wird deshalb auf Englisch angezeigt.

Überblick

MiniMax H3
Combat
BASE (V2) by FourBunny is a video
LoRA
focused on action: faster combat, better action continuity, believable hit reactions, cumulative damage, spatial movement and physical cause and effect. V2 goes beyond fights, adding richer body movement, finer sound detail and stronger shot-to-shot continuity to character performance.
Triggers: prfight1 for normal combat; add prfin1 after it for a clear finisher, knockout or takedown; add prslow1 for a brief slow-motion emphasis on the decisive strike. The author notes slow motion is not guaranteed.

Referenz

Name
Typ
Bedeutung
Trigger
prfight1
Normal combat scenes.
Finisher
prfin1
After prfight1 for a decisive ending.
Slow motion
prslow1
Brief emphasis on the final strike.
LoRA weight
0.7-1.0
Resolution
1080p / 2 MP
10 s clips take about 15 minutes on a 12 GB card per the author.
AD
Keine Installation nötig – KI-Modelle in der Cloud ausführen
BitVector Prism ist der einfachste Einstieg: Modell wählen, Prompt eingeben und in einer übersichtlichen Web-App generieren, ohne etwas einzurichten. BitVector gibt es auch auf Discord und im Web (SpyGlass).

Schritt für Schritt

  1. Describe the fighters, the space and who is winning in the integrated description.
  2. Put prfight1 at the start; add prfin1 if the clip ends the fight.
  3. Describe sound (impacts, breathing) in the soundscape.
  4. Weight 0.9, standard H3 steps.
  5. Keep each shot to one exchange of blows.

Beispiele

Sparring

prfight1. integrated_multimodal_description: [Shot 1] Two adult boxers spar in a dim gym. The taller one jabs, the shorter slips and lands a body shot; the taller staggers back into the ropes.
overall_soundscape: Glove impacts, squeaking shoes, heavy breathing.
non_diegetic_music: N/A

Finisher

prfight1, prfin1, prslow1. integrated_multimodal_description: [Shot 1] A swordswoman parries a strike and sweeps her opponent's legs; he falls and the blade stops at his collar.
overall_soundscape: Steel ring, boots on gravel, wind.

Creative idea: dialogue with tension

prfight1. integrated_multimodal_description: [Shot 1] Two negotiators face off across a table. One slams a hand down and says: <d>[English] Enough.</d> The other leans back slowly.
overall_soundscape: Hand on wood, chair creak.

Tipps

  • Name the hit reactions you want (staggers back, drops to one knee).
  • Cumulative damage: mention earlier hits in later shots.
  • Works for dialogue scenes too; it improves body language.
  • Combine with Better Motion at 0.4 for walking and turning.

Fehlerbehebung

Fight ends without impact

Warum es passiert
No finisher trigger.

So behebst du es
Add prfin1 after prfight1.

Limbs glitch in fast exchanges

Warum es passiert
Too many actions per shot.

So behebst du es
One exchange per shot; add Better Motion at 0.4.

AD
PirateDiffusion
Keine Installation nötig – KI-Modelle in der Cloud ausführen
PirateDiffusion gibt es nur auf Telegram und richtet sich an Profis: tausende Modelle, LoRAs und Workflows per Chat-Befehl, mit unbegrenzter Generierung zum Festpreis.

Fragen

Is slow motion reliable?

The author says prslow1 is not guaranteed; keep shots short.

Dialogue scenes?

Yes, V2 improves performance continuity in them.

Links und Quellen

Modelle in dieser Anleitung

Geschrieben von
Captain

Verwandte Anleitungen

Modelle
GalaxyAce LoRA for MiniMax H3: early-2010s low-end phone footage
aiguild's GalaxyAce LoRA recreates the look of a Samsung Galaxy Ace phone camera: muted colors, sensor noise, rolling shutter and everyday framing. Learn strengths per model, prompts and found-footage ideas.
Von Captain
2026-10-02
Modelle
Better Human Motion for MiniMax H3: natural, consistent body movement
AdaptiveVision's Better Motion LoRA makes MiniMax H3 (and LTX) characters move naturally and consistently. Learn weight, resolution, step count and how to write short motion prompts.
Von Captain
2026-10-01
Modelle
Passionate Kissing LoRA for MiniMax H3: convincing romantic scenes between adults
kermitfrog1202's LoRA improves kissing scenes in MiniMax H3 video. Learn the trigger sentence, the strength values the author found for different couples and how to stage a tasteful romantic shot.
Von Captain
2026-10-01
Modelle
MiniMax H3 Turbo LoRAs: 4-8 step video with the larryvrh pack
The MiniMax H3 Turbo LoRA pack (larryvrh 4-step and friends) cuts video generation steps from dozens to a handful. Learn the per-variant step counts, samplers and strengths, plus how to prompt H3 for fast, clean clips.
Von Captain
2026-10-01
Prompting
Prompting MiniMax H3: text to video, image to video, reference video and audio
How to direct MiniMax H3 from a chat command: the WHO + WHERE + ACTION + CAMERA + SOUND recipe for text to video, animating a still with image to video, giving every reference picture a job in reference mode, and getting clean dialogue instead of mumbling. With copy-and-try prompts and the author's PDF.
Von Mark White
2026-09-20

Weitere Anleitungen

Modelle
FLUX.1 Dev: how to prompt it, best settings and creative ideas
A practical guide to FLUX.1 Dev by Black Forest Labs: natural-language prompting, guidance and step settings, LoRA stacking, text rendering and prompt ideas that play to its strengths.
Von Captain
2026-09-30
Modelle
Stable Diffusion XL 1.0: prompts, settings and what it still does best
How to get the most out of the official SDXL 1.0 base model: native resolutions, CFG and sampler settings, the refiner, negative prompts and ideas for styles where SDXL still shines.
Von Captain
2026-10-01
Modelle
Stable Diffusion 1.5: the classic model, prompted properly
Stable Diffusion 1.5 is still worth knowing: the right resolution, CFG and sampler, how to use its huge library of LoRAs and embeddings, and creative prompt ideas suited to a 512-pixel model.
Von Captain
2026-10-02
Modelle
FLUX.2 Dev: prompting the newest Black Forest Labs model
FLUX.2 Dev brings larger prompts, better text, stronger realism and multi-reference editing. This guide covers the turbo workflow, guidance and resolution settings, prompt structure and ideas that use its new strengths.
Von Captain
2026-10-03
Modelle
Qwen-Image: long prompts, perfect text and bilingual posters
Qwen-Image is the model to reach for when the words in the picture matter. Learn how to write its long descriptive prompts, render English and Chinese text accurately, set CFG and steps, and explore layout-heavy creative ideas.
Von Captain
2026-09-29
Modelle
SDXL-Lightning: four-step generation on any SDXL checkpoint
ByteDance's SDXL-Lightning LoRA turns a 30-step SDXL render into a 4-step one. Learn the step counts, the CFG you must use, the sampler, and how to combine it with your favorite checkpoints and style LoRAs.
Von Captain
2026-09-30