A recipe for flawless one-step distant faces in H3 without a SAM face-swap pass
robertwellesley · reddit · 2026-11-04
A detailed community recipe for generating clear distant faces "in camera" in H3, avoiding an extra SAM-based face swap:
- 2-Stage H3 workflow with latent upscale; Singularity 2.0, Pika at 1.0, R34l1ty at 1.0, LMS 0.5, DDIM Uniform, 2 seeds at 10-12 steps, denoise 1.0.
- Reference image needs the face and side of head at 650-700px minimum plus a headless body, with "Match" enabled; follow the H3 Prompting Guide closely.
- Output is 4K (2x 1080p), downscaled back to 1080p, reliably yielding clear 60px-tall faces in a 1920x1080 scene.
- Verified for stills only (used for storyboarding); the author plans to test whether H3 preserves the face when the image feeds a video scene.
More from Multimodal
- Seedance 2.5 AI video of a wild Malfoy is turning heads — nikola_mr64990 · 2026-11-04
- Free course covers the 4 hard problems of production generative media agents — Al_Grigor · 2026-11-04
- PixelBench: testing multimodal models pixel-by-pixel instead of with QA — bdsqlsz · 2026-11-04
- Kitten-TTS 2 offers free voice cloning on laptops, no GPU needed — solyarisoftware · 2026-11-04
- "We'll Take Your RAM": a fully local-hardware cyberpunk meme from 2030 San Francisco — Wonderful-Yak-3348 · 2026-11-04
- Openfilm: open-source AI video agent turns coding agents into video makers — solyarisoftware · 2026-11-04