Minimax H3 I2V Test: Severe Temporal Noise and Low Resolution vs. Veo
Dumperandumper · reddit · 2026-09-01
A user tested Minimax H3's Image-to-Video capabilities and found significant temporal noise and low perceived resolution, especially in aerial shots, falling short of Google Veo 3.1. Despite experimenting with BF16, step counts, and quantization, the output retained a plastic look with long processing times. The user seeks advice on achieving broadcast-quality stability.
More from Multimodal
- MiniMax H3 Max supports reference images, generating 15s video in 17s — aziz4ai · 2026-09-01
- MiniMax H3 Struggles with Multi-Reference Character Replacement — webAd-8847 · 2026-09-01
- MiniMax H3 All-in-One Checkpoint Fuses Text, Image, and Turbo — AiCreatorCamp · 2026-09-01
- Fixing Dance Accuracy in Seedance 2.0 Using Depth Maps — jordiponsdotme · 2026-09-01
- Creative Prompt: 70 Years Unfold in One Room Following a Tortoise — umesh_ai · 2026-09-01
- Help: Best practices for tagging images when training Krea2 Style LoRA — BluetownA1 · 2026-09-01