Minimax H3 tip: blend latent noise to loosen overly literal reference images
neph1010 · reddit · 2026-09-16
A ComfyUI user shares a hands-on technique for Minimax H3's overly literal ref2va image references: instead of feeding the raw image, blend its latent with noise. Noise in latent space gives the sampler freedom to explore nearby concepts. Empirically, latent strength 0.35 is the sweet spot (keeps composition, regenerates texture); 0.38 stays blocky, 0.25 drifts toward no-reference output, and the drop-off is sharp. Since native nodes don't accept latent input, the author released custom modified nodes on GitHub; the same trick works well with Flux2 Klein's latent-as-reference.
More from Multimodal
- Midjourney dual --sref combo produces medieval woodcut cat knight art — sergeantsref · 2026-11-02
- Fluffy AI-generated character designs: fox girl, alien bug and cat photographer — OVolosin82152 · 2026-11-02
- Fable 5.5 reportedly one-shots animation in pure code as Anthropic's coding bet pays off — Hesamation · 2026-11-02
- FlowHMR gets 5x faster, demoing an LLM agent pipeline from video to photoreal mocap — camenduru · 2026-11-02
- Retro 1973 music video "Rock is Dead" by TECHHALLA showcases startlingly realistic AI generation — Mother-Pear-2392 · 2026-11-02
- Weekend splatting with COLMAP and Lichtfeld — janusch_patas · 2026-11-02