MiniMax H3 Experimental
w4a8 is a format using 4-bit weights and int8-convrot activations
Model source
- Kind
- Model
- File size
- 177 MB–21.3 GB
- Version
- vMinimax h3 4step lora flashgen v1.0 768p fl2va pruned avg rank 13 bf16
- Updated
- Oct 14, 2026
- Source
- Hugging Face
Source description
w4a8 is a format using 4-bit weights and int8-convrot activations
Requires ComfyUI 0.31.0
int8_convrot VAE needs ComfyUI 0.36.0 for maximum speed benefit
It speeds up VAE decode times by ~2x
ref lora is the difference between fl2va and ref2va, completely experimental, I don't even know if it has a use case at this point.
w6a8 is a format using 6-bit weights and int8-convrot activation, work in progress
minimax_h3_ref2va_pruned_w6a8_g32.safetensor
https://github.com/Comfy-Org/comfy-kitchen/pull/191
minimax_h3_lynnreal_light_vae_int8_convrot.safetensors
https://huggingface.co/stdstu123/LynnReal-Onmi-light-vae
int8-convrot quant of the pruned light VAE, this is about 1.3x faster for decoding.
PDMD loras
Converted and rank reduced from:
https://huggingface.co/pdmd2026/pdmd_2NFE_lora
https://huggingface.co/pdmd2026/pdmd_4NFE_lora
DMAD lora
Sources
1 source
Sources
1 sourceVerified Oct 14
Model artifacts
20 artifacts
Model artifacts
20 artifactsloras/minimax_h3_4step_lora_flashgen_v1.0_768p_fl2va_pruned_avg_rank_13_bf16.safetensors
safetensors · 177 MB · SHA-256 1254d62240b1…b067 · Hugging Face
Downloadloras/minimax_h3_DMAD_4step_full_lora_avg_rank_39_bf16.safetensors
safetensors · 350 MB · SHA-256 ddfdd94f4c9a…2d11 · Hugging Face
Downloadloras/minimax_h3_ELM_longlive_plug_4step_lora_bf16.safetensors
safetensors · 1.82 GB · SHA-256 4a1058d29eec…f214 · Hugging Face
Downloadloras/minimax_h3_pdmd_2step_lora_avg_rank_38_bf16.safetensors
safetensors · 363 MB · SHA-256 a17f3eb45ab1…ec55 · Hugging Face
Downloadloras/minimax_h3_pdmd_4step_lora_avg_rank_57_bf16.safetensors
safetensors · 533 MB · SHA-256 b5c62eb228a4…fc37 · Hugging Face
Downloadloras/minimax_h3_ref_lora_rank_256_bf16.safetensors
safetensors · 2.40 GB · SHA-256 7979fb9e5ac8…9c87 · Hugging Face
Downloadloras/MiniMax-H3-FL2VA-Acc-8Step_comfy.safetensors
safetensors · 1.61 GB · SHA-256 71347e272513…71ac · Hugging Face
Downloadloras/MiniMax-H3-FL2VA-Acc-8Step_pruned_comfy.safetensors
safetensors · 1.61 GB · SHA-256 e97b813a6f85…a897 · Hugging Face
Downloadloras/MiniMax-H3-Ref2VA-Acc-8Step_comfy.safetensors
safetensors · 1.61 GB · SHA-256 d7cd9480c000…7460 · Hugging Face
Downloadloras/MiniMax-H3-Ref2VA-Acc-8Step_pruned_comfy.safetensors
safetensors · 1.61 GB · SHA-256 6f18e1c2eccb…1e39 · Hugging Face
Downloadminimax_h3_fastvideo_vsa_datafree_1300step_4step_int8_convrot.safetensors
safetensors · 21.3 GB · SHA-256 7221ae65d787…13d3 · Hugging Face
Downloadminimax_h3_fl2va_pruned_w4a8_mixed.safetensors
safetensors · 11.7 GB · SHA-256 01aa7b92c007…8dec · Hugging Face
Downloadminimax_h3_lynnreal_light_vae_int8_convrot.safetensors
safetensors · 1.99 GB · SHA-256 f11f8b9b96c9…e7e2 · Hugging Face
Downloadminimax_h3_ref2va_pruned_w4a8_mixed.safetensors
safetensors · 11.0 GB · SHA-256 de2c6c29c4ee…9910 · Hugging Face
Downloadminimax_h3_video_vae_int8_convrot.safetensors
safetensors · 2.95 GB · SHA-256 9bb2d96f218c…3410 · Hugging Face
Downloadmodel_patches/minimax_h3_fun_controlnet_union_2.0_pruned_bf16.safetensors
safetensors · 7.81 GB · SHA-256 cd48bdc11b1f…82b4 · Hugging Face
Downloadmodel_patches/minimax_h3_fun_controlnet_union_2.0_pruned_int8_convrot.safetensors
safetensors · 4.22 GB · SHA-256 890af58bb350…c612 · Hugging Face
Downloadmodel_patches/minimax_h3_fun_controlnet_union_pruned_bf16.safetensors
safetensors · 3.93 GB · SHA-256 57fe1e64928a…1c48 · Hugging Face
Downloadmodel_patches/minimax_h3_fun_controlnet_union_pruned_int8_convrot.safetensors
safetensors · 2.14 GB · SHA-256 9c645c0a308c…ed0c · Hugging Face
Downloadwip/minimax_h3_ref2va_pruned_w6a8_g32.safetensors
safetensors · 15.6 GB · SHA-256 565519608d06…db01 · Hugging Face
DownloadSource excerpts
2 excerpts
Source excerpts
2 excerptsw4a8 is a format using 4-bit weights and int8-convrot activations Requires ComfyUI 0.31.0 --- int8_convrot VAE needs ComfyUI 0.36.0 for maximum speed benefit It speeds up VAE decode times by ~2x --- ref lora is the difference between fl2va and ref2va, completely experimental, I don't even know if it has a use case at this point. --- w6a8 is a format using 6-bit weights and int8-convrot activation, work in progress ## minimax_h3_ref2va_pruned_w6a8_g32.safetensor https://github.com/Comfy-Org/comfy-kitchen/pull/191 --- ## minimax_h3_lynnreal_light_vae_int8_convrot.safetensors https://huggingface.co/stdstu123/LynnReal-Onmi-light-vae int8-convrot quant of the pruned light VAE, this is about 1.3x faster for decoding. --- ## PDMD loras Converted and rank reduced from: https://huggingface.co/pdmd2026/pdmd_2NFE_lora https://huggingface.co/pdmd2026/pdmd_4NFE_lora ## DMAD lora https://huggingface.co/ZhengmingYu/DMAD
Source context: 0 downloads · 593 likes · Repo Kijai/MiniMax-H3-experimental
