MiniMax H3 Character-Swap LoRA Runs on 8GB VRAM, Tested on RTX 4070
big-boss_97 · reddit · 2026-10-08
A Reddit user tested the community Character Swap LoRA for MiniMax H3 on a local RTX 4070 with 8GB VRAM and 64GB RAM, generating an 832×640 video in about 12:45 and changing facial expressions via prompt.
The LoRA weights are available on Hugging Face (akatz-ai/MiniMax-H3-Character-Swap-LoRA), with a YouTube tutorial for low-VRAM users to reproduce locally.
More from Multimodal
- Codex+Tripo toolkit fixes AI 3D models: agent-assisted refinement loop — tinyfool · 2026-10-09
- Testing LTX 2.5 SDR-to-HDR conversion: results better than expected — ShroakzGaming · 2026-10-09
- Dev streams a 66GB unquantized 22B video model on an iGPU with only 15.6GB shared RAM — Business_Swordfish_5 · 2026-10-09
- Qwen-Image-2.1 Outfit-Swap Consistency LoRA Trends on Hugging Face — ausboss · 2026-10-09
- How Seedance V2 shocked the industry: 1,000+ data specialists and licensed cinema-grade footage, not scraped Douyin — creatoroff · 2026-10-09
- Running MiniMax H3 on a single RTX 5090: 0.8MP is the ceiling for 15s clips — Realistic-Fennel-190 · 2026-10-09