MMH3 Realism Slider V2 — HyperFlow H3… V2 - Base H3

MMH3 Realism Slider V2 — HyperFlow H3 Flash & MiniMax H3 - V2 - Base H3

NSFW

A skin realism LoRA for HyperFlow H3 Flash (ref2va and fl2va) and MiniMax H3 video: less plastic shine, more natural skin.

What's new in V2

V1 was a block-renumbered port of @alcaitiff's original rank-1, MLP-only slider (https://civarchive.com/models/3014078). 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 (qkv_proj, out_proj) and MLP (fc1, fc2) in H3's later blocks (100 layers):

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.
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.

Which file to download

V2 - ref2va: for minimax_h3_ref2va_hyperflow_flash_14b with ComfyUI's stock LoraLoaderModelOnly. Strength 0.65 to 0.85, default 0.75.
V2 - fl2va: for minimax_h3_fl2va_hyperflow_flash_14b 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.
V2 - Base H3: the file for H3 Editor (comfyui-h3-editor, https://github.com/igorls/comfyui-h3-editor) on any checkpoint, at strength 1.2 with start_sigma 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.

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.)

Description

V2 for the 50-block MiniMax H3, and the one file to use with H3 Editor on any checkpoint

A new rank-16 LoRA that trains both self-attention and MLP on blocks 24 to 48 (100 layers). Unlike V1 (rank 1, MLP only), it changes skin on smooth reference portraits in ref2va too: it takes the plastic shine off and brings back matte, natural skin texture, and is trained to leave eyes, eyelashes, eyebrows and stubble alone.

H3 Editor (comfyui-h3-editor, https://github.com/igorls/comfyui-h3-editor): works on MiniMax H3 and on both HyperFlow H3 Flash models. The H3 LoRA node renumbers the blocks HyperFlow dropped and adds the LoRA at run time on int8 and factorized layers. Strength 1.0 to 1.4, default 1.2, with start_sigma 0.75, which leaves the first steps alone so pose and framing stay close to the render without the LoRA. For a reference with large wet highlights on the cheeks or forehead, try start_sigma 1.0 at strength 0.75. The start_sigma input is in H3 Editor 0.3.0 and later, where finals use this file by default.
Stock ComfyUI (LoraLoaderModelOnly): on the 50-block MiniMax H3, strength 0.65 to 0.85. On MiniMax H3 Image, which has its own image VAE, the effect is gentler unless pushed to 1.2 to 1.6.