Sorti le même jour que Seedance 2.5, Minimax H3, la nouvelle IA Vidéo de Hailuo se dote du son.
Les specs : 2K (1440 sur le petit côté), 24 fps, des clips de 5 à 15 secondes, et surtout du son généré dans la même passe que l’image, dialogues compris.
Le système de références accepte 9 images, 3 vidéos et 3 audios en même temps, ce qui permet de tenir un visage, un mouvement et une voix sur plusieurs plans.
Sur les classements Artificial Analysis, il prend la 1re place en édition vidéo, la 2e en text-to-video et la 3e en image-to-video.
Côté accès : dispo sur Hailuo AI, sur l’app desktop MiniMax Hub, et déjà sur FAL avec trois endpoints (texte, image, référence). Tarif annoncé à 0,13 $ la seconde en 2K, avec un palier 768p moins cher à venir.
MiniMax annonce aussi une sortie des poids en open weight « dans les prochains jours ». À l’heure où j’écris, rien n’est publié sur leur Hugging Face. À suivre.
Mon test, retour à Ithaque en 15 secondes avec 4 photos de référence :
Le prompt :
REFERENCES:
- Image 1 = the hero’s identity (character sheet, multiple angles). Hold his
face, build and outfit details from this reference in every shot.- Image 2 = the wife’s identity. Woman in her 40s, shoulder-length
copper-blonde hair, warm brown eyes, hoop earrings, calm expression.- Image 3 = the son’s identity. Boy around 10, short dark brown hair spiked
up, tanned skin, bright smile.- Image 4 = the daughter’s identity. Teenage girl around 15, dark brown hair
in two braids, wide smile, expressive eyes.
Hold all four faces consistent across the entire clip.CHARACTERS:
- ULYSSE (Image 1): man in his forties, weathered from years at sea. Off-white
linen tunic with a leather chest strap, bronze arm guards, a dust-red cloak,
leather sandals. On his head, a burgundy baseball cap with a white « NY »
embroidered on the front — he wears it in every single shot, it never comes
off. Salt-stained skin, three days of stubble, tired but shining eyes.- PÉNÉLOPE (Image 2): long ivory Greek chiton with a woven blue belt, hair
loose over her shoulders, a simple gold pendant. Standing very still.- TÉLÉMAQUE (Image 3): short sand-coloured tunic, leather sandals, a small
wooden sword hanging at his belt.- ATHÉNA (Image 4): flowing white and gold robe, polished bronze breastplate,
golden laurel crown over her braids, a tall spear in her right hand. Serene,
almost divine composure.SETTING: a rocky bay on the island of Ithaca at golden hour. Turquoise water,
pale limestone cliffs, silver olive trees on the slopes, worn stone steps
climbing from the beach toward white stone houses. Warm low sun, long shadows,
dust and sea spray catching the light. Photorealistic epic cinema,
Mediterranean colour grading — warm gold, deep blue shadows.0–3s — The Return
A wooden Greek galley glides into the bay, square sail half-furled, oars
lifting. ULYSSE stands alone at the prow, one hand on the rigging, scanning
the shore. The wind pushes his cloak back. He recognises the coastline and his
jaw tightens.
Camera: slow wide tracking shot alongside the ship, then push in to a medium
shot on his face at the prow.3–5s — The Landing
He climbs over the rail and drops into the shallow water, wading toward the
beach with heavy steps, water up to his knees. He stops on the wet sand, looks
up at the stone steps.
Camera: low angle from the waterline, following him toward shore.5–8s — The Family
At the top of the steps, three silhouettes appear against the sunset:
PÉNÉLOPE in the centre, ATHÉNA to her right holding her spear, TÉLÉMAQUE
half-hidden behind his mother. A long beat of silence. Then the boy breaks
away and runs down the steps.
Camera: reverse shot from behind Ulysse, revealing the three figures, then cut
to a medium shot on the family.8–11s — Télémaque
The boy runs full speed across the sand and throws himself into his father’s
arms. Ulysse catches him and lifts him off the ground, laughing, eyes closed.
TÉLÉMAQUE (French, child’s voice, breathless with joy): « Père ! Tu es
revenu ! »
ULYSSE (French, low and hoarse, close to tears): « J’ai traversé toutes les
mers pour vous. »
Camera: handheld medium shot circling the two of them.11–13s — Pénélope
She walks down the last steps and stops in front of him. She lifts a hand to
his face, slowly, as if checking he is real.
PÉNÉLOPE (French, quiet, steady, twenty years of waiting in her voice):
« Vingt ans, Ulysse. Vingt ans. »
He covers her hand with his own. They embrace, the boy still clinging to his
side.
Camera: slow push-in to a tight two-shot.13–15s — Athéna
ATHÉNA descends last, plants her spear in the sand and smiles at her father.
ATHÉNA (French, warm and clear): « Papa, bienvenue chez toi. »
Camera pulls back and rises into a wide aerial shot: the four of them together
on the beach, the galley anchored in the bay, olive slopes and white houses
behind, the sun touching the horizon. Freeze on the wide.AUDIO (native stereo, generated with the picture):
- Ambience: waves breaking on rock, wind in the rigging, wood creaking, gulls
in the distance, sandals on wet sand.- Foley: the splash as he drops into the water, the boy’s running footsteps,
the spear butt striking the sand at 13s.- Dialogue: all lines in French, intimate and close-miked, never shouted.
- Score: a single low string drone from 0s, a solo lyre entering at 8s, full
warm strings swelling from 11s and resolving on the wide shot at 14s.- Keep the score silent between 5s and 8s — only wind and waves during the
moment of recognition.CONSTRAINTS:
- 16:9, photorealistic epic cinema throughout, no stylisation or animation.
- All four faces fully visible and consistent with Images 1 to 4.
- The burgundy NY cap stays on the hero’s head in every shot, including the
landing in the water and the embrace.- No modern objects, no brand names, no logos anywhere in frame apart from the
cap.- Natural performances, no theatrical over-acting.
La ressemblance tient, le son est vraiment au niveau. Le bémol : watermark si tu n’as pas d’abonnement.
Si vous testez, partagez vos réglages de références, je pense qu’il y a beaucoup à gagner là-dessus.