A skin realism LoRA for HyperFlow H3 Flash (ref2va and fl2va) and MiniMax H3 video: less plastic shine, more natural skin.<h3 id="what's-new-in-v2">What's new in V2</h3>V1 was a block-renumbered port of @alcaitiff's original rank-1, MLP-only slider (<a target="_blank" rel="ugc" href="https://civarchive.com/models/3014078">https://civarchive.com/models/3014078</a>). Because V1 only touched MLP layers, smooth reference portraits in ref2va passed through it by way of self-attention and barely changed.V2 is trained from scratch at rank 16 on both self-attention (<code>qkv_proj</code>, <code>out_proj</code>) and MLP (<code>fc1</code>, <code>fc2</code>) in H3's later blocks (100 layers):<ul><li>It changes smooth reference skin in ref2va. It takes off plastic highlights and brings back natural skin texture. We tested it on fair, olive, East Asian, South Asian and deep-melanin skin, and did not see added grain or ageing.</li><li>It is trained to leave fine features alone. Eyes, eyelashes, irises, eyebrows, beards and hair stayed sharp in our tests, in motion and in 1080p finals.</li></ul><h3 id="which-file-to-download">Which file to download</h3><ul><li>V2 - ref2va: for <code>minimax_h3_ref2va_hyperflow_flash_14b</code> with ComfyUI's stock <code>LoraLoaderModelOnly</code>. Strength 0.65 to 0.85, default 0.75.</li><li>V2 - fl2va: for <code>minimax_h3_fl2va_hyperflow_flash_14b</code> with the stock loader. Strength 0.65 to 0.90, default 0.80. 97 of its 100 layers load there; the warning about the other 3 is expected.</li><li>V2 - Base H3: the file for H3 Editor (<code>comfyui-h3-editor</code>, <a target="_blank" rel="ugc" href="https://github.com/igorls/comfyui-h3-editor">https://github.com/igorls/comfyui-h3-editor</a>) on any checkpoint, at strength 1.2 with <code>start_sigma</code> 0.75, and for the stock loader on the unpruned 50-block MiniMax H3 at 0.65 to 0.85. With H3 Editor, use this file even on HyperFlow: its H3 LoRA node renumbers the blocks itself.</li></ul>No trigger words.The comparison clips are short, one to two seconds each, and show the same shot with and without the LoRA.(The V1 tabs stay available. Credit to @alcaitiff for the original V1 slider.)

MMH3 Realism Slider V2 — HyperFlow H3 Flash & MiniMax H3

V2 for <code>minimax_h3_ref2va_hyperflow_flash_14b</code> (reference pictures and first/last-frame video), for ComfyUI's stock LoRA loaderA new rank-16 LoRA, numbered for the 49-block HyperFlow ref2va model (which drops H3 block 9). V1 only touched the MLP layers, so smooth reference portraits passed straight through it by way of self-attention. V2 trains both self-attention and MLP on blocks 23 to 47 (100 layers), to take the plastic shine off skin and bring back natural skin texture while leaving eyelashes, irises and stubble alone.<ul><li>Stock ComfyUI (<code>LoraLoaderModelOnly</code>): strength 0.65 to 0.85, default 0.75. All 100 layers load.</li><li>H3 Editor (<code>comfyui-h3-editor</code>): download V2 - Base H3 instead. The editor's H3 LoRA node renumbers the base file for HyperFlow by itself.</li></ul>

Files with SHA256: 730c8bc08184297c70b8225865778828bc935f3e1ff8bd2d7a382c737ab7874f