Local pipeline showcase: MiniMax H3 refMod generations from reference image to 4K upscale
edisson75 · reddit · 2026-11-13
Reddit user edisson75 shared results from a fully local image generation pipeline built around MiniMax H3 refMod: reference image → Qwen3.8 27B (HauhauCS Q2KP quant) for prompt generation → mmH3 refMod generation → 4x latent upscale → second pass → mmH3 16x image upscale.
The attached result shows near-commercial portrait quality with fine facial detail and lighting, demonstrating how quantized local model chains can handle reference-based character generation. The poster is asking the community to judge whether the output quality is good enough.
More from Multimodal
- Engineer builds 3D Ayodhya model with Chisel, shares the file publicly — gowthami_s · 2026-11-13
- 67B MiniMax H3-Image runs on a 16GB MacBook Air via MLX with ~85% less peak memory — FlimsyBrief4232 · 2026-11-13
- Viral image prompt: watch the alien invasion from your penthouse — charis_ai · 2026-11-13
- Ant Group's Duo v1: a reciprocal world model for real-time full-duplex interaction — inclusionAI · 2026-11-13
- G-STAR from PKU & Meituan LongCat: 20x faster autoregressive image generation at 2.5s per image — meituan-longcat · 2026-11-13
- Dev builds ComfyUI nodes to add batch generation to MiniMax H3 workflows — an80sPWNstar · 2026-11-13