Testing MiniMax-H3 to LTX 2.5 Upscaling Workflow: Promising Results
iiTzMYUNG · reddit · 2026-09-01
The author tested a workflow combining MiniMax-H3 video generation with LTX 2.5 upscaling. Results show that better MiniMax-H3 inputs lead to better upscales, though face consistency still needs stronger input generation. On an RTX 3060 12GB, upscaling takes 15 mins at 0.6 resolution and 20-30 mins at 0.8-1.0. The author plans to release the workflow after further optimization.
More from Multimodal
- Quibble: MiniMax H3 experiment for persistent character animation in ComfyUI — magratheya_64 · 2026-09-01
- LibTV Adds Depth-Video Extraction: Copy Motion and Camera Work, Not Faces — 卡尔的AI沃茨 · 2026-09-01
- Funny Dinosaur Safari Image Generated by Dreamina — umesh_ai · 2026-09-01
- Gemini Omni Flash 1.1 supports start/end frame generation with cost optimization — azed_ai · 2026-09-01
- Free local 3D world pipeline: shooting a car commercial inside an AI world — mickmumpitz · 2026-09-01
- Creator Showcases AI-Generated Video in 1930s Animation Style — Secret_Wasabi_2373 · 2026-09-01