MiniMax H3 Dual-Sampling Workflow: Latent Upscaling & Audio Restoration Guide
wjc_5 · reddit · 2026-08-23
The author tested an updated MiniMax H3 workflow to address issues like limited upscale factors, waxy live-action results, weak audio, and high VRAM usage.
- Latent Upscaling Model: Replacing basic latent resize with a model-based latent upscale eliminates line-like artifacts and fragmented shapes after the second pass. A 1.0x alignment step is recommended for non-integer factors.
- Dual-Sampling Setup: Using the 8-step LoRA for the first pass and 4-step LoRA for the second pass preserves high-motion clothing edges better and avoids character duplication compared to using the 4-step LoRA too early.
- Visuals & Audio: Lowering LoRA strengths (0.75 for 8-step, 0.7 for 4-step) reduces the waxy look in live-action videos. Adding a separate voice-reference audio clip improves speech quality.
- Sampling: Euler with the Beta scheduler works well.
Note that while dual sampling speeds up the first part, it does not reduce peak hardware requirements. A detailed tutorial is available on YouTube.
Related event: MiniMax H3 Tuning Guide: LoRA, Samplers and Speed-Quality Tradeoffs(4 posts)→
More from Multimodal
- AssemblyAI's Universal-3.6 Pro speech-to-text lands on OpenRouter at $0.30/hr — AssemblyAI · 2026-10-16
- FLUX 3 Image Klein lands on fal: box-drawn layouts, multi-reference placement, style transfer — bfl_ai · 2026-10-16
- SGLang ships Day-0 support for FLUX 3 Image Klein: 10 reference images, BF16/FP8/NVFP4 variants — ying11231 · 2026-10-16
- Krea 3 Scores 4.81 on CPI-Bench, Second Only to GPT Image 2.5 — testingcatalog · 2026-10-16
- Higgsfield teases Kling 4.0 rollout with a monkey-driven demo video — azed_ai · 2026-10-16
- Halcyon early-fusion image models match late-fusion with 275x less pretraining compute, open-sourced — davidwromero · 2026-10-16