Instructions to use Alex995647/loras-minimax-h3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Alex995647/loras-minimax-h3 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("MiniMaxAI/MiniMax-H3", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("Alex995647/loras-minimax-h3") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
Add LightX2V Turbo REF2V 4-Step v0.1 (resized rank 20)
Browse files- lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20/example_images/NO_PREVIEWS.txt +1 -0
- lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20/info.txt +140 -0
- lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20/metadata.json +93 -0
- lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20/minimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors +3 -0
lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20/example_images/NO_PREVIEWS.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
The repo Kijai/MiniMax-H3_comfy ships no sample images or videos.
|
lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20/info.txt
ADDED
|
@@ -0,0 +1,140 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
═══════════════════════════════════════════════════════════════════════════════
|
| 2 |
+
LIGHTX2V TURBO REF2V 4-STEP V0.1 (RESIZED RANK 20)
|
| 3 |
+
═══════════════════════════════════════════════════════════════════════════════
|
| 4 |
+
|
| 5 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 6 |
+
IDENTITY
|
| 7 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 8 |
+
Base model MiniMax H3
|
| 9 |
+
Generation mode Reference-to-Video (Ref2VA), First-and-Last-Frame-to-Video (FL2VA)
|
| 10 |
+
Checkpoint ref2va (drives Ref2VA only)
|
| 11 |
+
Category Speed / step-reduction
|
| 12 |
+
Author Kijai
|
| 13 |
+
Source https://huggingface.co/Kijai/MiniMax-H3_comfy
|
| 14 |
+
Provenance HuggingFace · Kijai/MiniMax-H3_comfy · loras/minimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors
|
| 15 |
+
|
| 16 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 17 |
+
WEIGHTS
|
| 18 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 19 |
+
File minimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors
|
| 20 |
+
Version main
|
| 21 |
+
Size 292.5 MB
|
| 22 |
+
Format SafeTensor / unknown precision
|
| 23 |
+
SHA256 9ea3bd3a6aac22994153e294cf1ecab0a8766fc0f8d056ace645a01d1a6a4daf
|
| 24 |
+
Published 2026-08-13
|
| 25 |
+
|
| 26 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 27 |
+
TRIGGER WORDS
|
| 28 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 29 |
+
No trigger word. The LoRA applies as soon as it is loaded —
|
| 30 |
+
strength is the only way to dial the effect up or down.
|
| 31 |
+
|
| 32 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 33 |
+
RECOMMENDED SETTINGS
|
| 34 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 35 |
+
Taken verbatim from the author's description:
|
| 36 |
+
|
| 37 |
+
LoRA strength 0.75
|
| 38 |
+
Sampler steps 4
|
| 39 |
+
|
| 40 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 41 |
+
ADOPTION & POPULARITY
|
| 42 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 43 |
+
Downloads 0
|
| 44 |
+
Rating 100% positive (412/412)
|
| 45 |
+
Comments 0
|
| 46 |
+
Published 22 days ago
|
| 47 |
+
Download rate 0.0/day since release
|
| 48 |
+
|
| 49 |
+
Adoption rank #94 of 114 in this collection (Niche)
|
| 50 |
+
Momentum rank #94 of 114 by download rate (Long tail · fresh release)
|
| 51 |
+
|
| 52 |
+
Adoption tier is the percentile of total downloads within this collection;
|
| 53 |
+
momentum is the percentile of downloads-per-day since release. Momentum is
|
| 54 |
+
ranked rather than measured against a fixed rate, because this base model
|
| 55 |
+
is itself new and every LoRA looks fast on an absolute scale. Both numbers
|
| 56 |
+
are CivitAI's, read on the scrape date at the foot of this file.
|
| 57 |
+
|
| 58 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 59 |
+
CHAINING / STACKING
|
| 60 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 61 |
+
Stacks with others not stated
|
| 62 |
+
|
| 63 |
+
The author gave no chaining guidance. The rules below are the
|
| 64 |
+
general MiniMax H3 ones, not anything specific to this LoRA.
|
| 65 |
+
|
| 66 |
+
General MiniMax H3 rules that apply here (see REFERENCE.txt):
|
| 67 |
+
|
| 68 |
+
• Load the LoRA with LoraLoaderModelOnly between UNETLoader and your
|
| 69 |
+
sampler; stacking means chaining several of those nodes in series.
|
| 70 |
+
• This is a step-distillation (turbo) LoRA. DISTILL LORAS DO NOT COMPOSE
|
| 71 |
+
— running this alongside another turbo/PDD/lightx2v LoRA breaks
|
| 72 |
+
generation. Pick exactly one.
|
| 73 |
+
• Style and character LoRAs stack on top of it normally, with no special
|
| 74 |
+
strength penalty beyond ordinary multi-LoRA tuning.
|
| 75 |
+
• Use Euler so the trained sigma schedule is consumed as intended, and
|
| 76 |
+
keep CFG at 1.0 with BasicGuider.
|
| 77 |
+
• Only the step counts the LoRA was trained for (usually 4, 6 or 8) work.
|
| 78 |
+
Other values produce noise — it is not a free dial.
|
| 79 |
+
• Chain this only onto the ref2va checkpoint. Loading it against the
|
| 80 |
+
fl2va checkpoint can degrade output silently rather than erroring.
|
| 81 |
+
|
| 82 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 83 |
+
PROS
|
| 84 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 85 |
+
+ Positively received — 412 thumbs-up with no down-votes. [source:
|
| 86 |
+
CivitAI stats]
|
| 87 |
+
+ Author documents use across several modes (REF2V, FL2V). [source:
|
| 88 |
+
description]
|
| 89 |
+
+ Compact at 293 MB — cheap to stack with other LoRAs. [source: file
|
| 90 |
+
size]
|
| 91 |
+
+ Author published concrete recommended settings. [source: description]
|
| 92 |
+
|
| 93 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 94 |
+
CONS & CAVEATS
|
| 95 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 96 |
+
- No trigger word — the LoRA is always on once loaded, so strength is the
|
| 97 |
+
only control. [source: model metadata]
|
| 98 |
+
- Modest adoption (0 downloads) — less community feedback to rely on.
|
| 99 |
+
[source: CivitAI stats]
|
| 100 |
+
|
| 101 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 102 |
+
COMPATIBILITY WARNING
|
| 103 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 104 |
+
MiniMax H3 ships two transformers that share tensor key names: the fl2va
|
| 105 |
+
checkpoint (driving T2VA, I2VA, L2VA and FL2VA) and the separate ref2va
|
| 106 |
+
checkpoint (driving Ref2VA). Loading a LoRA against the wrong family can
|
| 107 |
+
silently degrade output instead of raising an error, so confirm the family
|
| 108 |
+
above matches the checkpoint you have loaded. Likewise, LoRAs built for
|
| 109 |
+
full checkpoints often throw adaln_proj shape errors on pruned ones — match
|
| 110 |
+
the build to your checkpoint.
|
| 111 |
+
|
| 112 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 113 |
+
AUTHOR'S DESCRIPTION
|
| 114 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 115 |
+
ComfyUI conversion of LightX2V's 4-step distillation LoRA for the REF2V
|
| 116 |
+
(reference-to-video) MiniMax-H3 checkpoint, dynamically resized from rank
|
| 117 |
+
128 to an average rank of about 20 in BF16 (header metadata: sv_fro 0.96
|
| 118 |
+
dynamic resize from 128, alpha 8, scale 0.0625 baked in as a flat
|
| 119 |
+
multiplier). Load it on the REF2V checkpoint rather than the FL2V one; the
|
| 120 |
+
README's demo settings are 4 steps at 0.75 strength.
|
| 121 |
+
|
| 122 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 123 |
+
EXAMPLE MEDIA
|
| 124 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 125 |
+
No work-safe previews were published for this version.
|
| 126 |
+
|
| 127 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 128 |
+
HOW TO USE (COMFYUI)
|
| 129 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 130 |
+
1. Copy the .safetensors file into ComfyUI/models/loras/
|
| 131 |
+
2. Load the matching H3 checkpoint with UNETLoader.
|
| 132 |
+
3. Insert LoraLoaderModelOnly between the loader and your sampler.
|
| 133 |
+
4. Set strength as listed above, then add any trigger word to the prompt.
|
| 134 |
+
5. Keep CFG at 1.0 with BasicGuider unless you know you want real CFG.
|
| 135 |
+
|
| 136 |
+
───────────────────────────────────────────────────────────────────────────────
|
| 137 |
+
Scraped by lorakit on 2026-09-04 23:06 UTC.
|
| 138 |
+
Stats and description are the author's; pros/cons are derived from the
|
| 139 |
+
evidence tagged beside each line.
|
| 140 |
+
───────────────────────────────────────────────────────────────────────────────
|
lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20/metadata.json
ADDED
|
@@ -0,0 +1,93 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"name": "LightX2V Turbo REF2V 4-Step v0.1 (resized rank 20)",
|
| 3 |
+
"slug": "lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20",
|
| 4 |
+
"base_model": "MiniMax H3",
|
| 5 |
+
"media_type": "video",
|
| 6 |
+
"model_id": 1792216682,
|
| 7 |
+
"version_id": 1792216683,
|
| 8 |
+
"version_name": "main",
|
| 9 |
+
"source": "huggingface",
|
| 10 |
+
"url": "https://huggingface.co/Kijai/MiniMax-H3_comfy",
|
| 11 |
+
"repo_id": "Kijai/MiniMax-H3_comfy",
|
| 12 |
+
"file_path": "loras/minimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors",
|
| 13 |
+
"author": "Kijai",
|
| 14 |
+
"category": {
|
| 15 |
+
"key": "accelerator",
|
| 16 |
+
"label": "Speed / step-reduction"
|
| 17 |
+
},
|
| 18 |
+
"modalities": [
|
| 19 |
+
"ref2v",
|
| 20 |
+
"fl2v"
|
| 21 |
+
],
|
| 22 |
+
"primary_modality": "ref2v",
|
| 23 |
+
"checkpoint_family": "ref2va",
|
| 24 |
+
"trigger_words": [],
|
| 25 |
+
"recommended_settings": {
|
| 26 |
+
"LoRA strength": "0.75",
|
| 27 |
+
"Sampler steps": "4"
|
| 28 |
+
},
|
| 29 |
+
"chaining": {
|
| 30 |
+
"author": {
|
| 31 |
+
"stacks": "not stated",
|
| 32 |
+
"notes": "",
|
| 33 |
+
"avoid": [],
|
| 34 |
+
"requires": [],
|
| 35 |
+
"order": "",
|
| 36 |
+
"strength": "",
|
| 37 |
+
"stated": false
|
| 38 |
+
},
|
| 39 |
+
"general_rules": [
|
| 40 |
+
"Load the LoRA with LoraLoaderModelOnly between UNETLoader and your sampler; stacking means chaining several of those nodes in series.",
|
| 41 |
+
"This is a step-distillation (turbo) LoRA. DISTILL LORAS DO NOT COMPOSE — running this alongside another turbo/PDD/lightx2v LoRA breaks generation. Pick exactly one.",
|
| 42 |
+
"Style and character LoRAs stack on top of it normally, with no special strength penalty beyond ordinary multi-LoRA tuning.",
|
| 43 |
+
"Use Euler so the trained sigma schedule is consumed as intended, and keep CFG at 1.0 with BasicGuider.",
|
| 44 |
+
"Only the step counts the LoRA was trained for (usually 4, 6 or 8) work. Other values produce noise — it is not a free dial.",
|
| 45 |
+
"Chain this only onto the ref2va checkpoint. Loading it against the fl2va checkpoint can degrade output silently rather than erroring."
|
| 46 |
+
]
|
| 47 |
+
},
|
| 48 |
+
"pros": [
|
| 49 |
+
"Positively received — 412 thumbs-up with no down-votes. [source: CivitAI stats]",
|
| 50 |
+
"Author documents use across several modes (REF2V, FL2V). [source: description]",
|
| 51 |
+
"Compact at 293 MB — cheap to stack with other LoRAs. [source: file size]",
|
| 52 |
+
"Author published concrete recommended settings. [source: description]"
|
| 53 |
+
],
|
| 54 |
+
"cons": [
|
| 55 |
+
"No trigger word — the LoRA is always on once loaded, so strength is the only control. [source: model metadata]",
|
| 56 |
+
"Modest adoption (0 downloads) — less community feedback to rely on. [source: CivitAI stats]"
|
| 57 |
+
],
|
| 58 |
+
"file": {
|
| 59 |
+
"name": "minimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors",
|
| 60 |
+
"size_bytes": 306731560,
|
| 61 |
+
"size_mb": 292.52,
|
| 62 |
+
"sha256": "9ea3bd3a6aac22994153e294cf1ecab0a8766fc0f8d056ace645a01d1a6a4daf"
|
| 63 |
+
},
|
| 64 |
+
"stats": {
|
| 65 |
+
"downloads": 0,
|
| 66 |
+
"thumbs_up": 412,
|
| 67 |
+
"thumbs_down": 0
|
| 68 |
+
},
|
| 69 |
+
"popularity": {
|
| 70 |
+
"downloads": 0,
|
| 71 |
+
"thumbs_up": 412,
|
| 72 |
+
"thumbs_down": 0,
|
| 73 |
+
"comments": 0,
|
| 74 |
+
"approval": "100% positive (412/412)",
|
| 75 |
+
"age_days": 22.0,
|
| 76 |
+
"downloads_per_day": 0.0,
|
| 77 |
+
"rank_by_downloads": 94,
|
| 78 |
+
"rank_by_momentum": 94,
|
| 79 |
+
"of_total": 114,
|
| 80 |
+
"adoption_tier": "Niche",
|
| 81 |
+
"momentum": "Long tail · fresh release"
|
| 82 |
+
},
|
| 83 |
+
"selection_score": 4944.0,
|
| 84 |
+
"selection_reasons": [
|
| 85 |
+
"0 downloads",
|
| 86 |
+
"412 thumbs-up",
|
| 87 |
+
"top-ranked for its category"
|
| 88 |
+
],
|
| 89 |
+
"example_media": [],
|
| 90 |
+
"tags": [],
|
| 91 |
+
"scraped_at": "2026-09-04T23:06:39+00:00",
|
| 92 |
+
"path": "minimax-h3/lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20"
|
| 93 |
+
}
|
lightx2v-turbo-ref2v-4-step-v0-1-resized-rank-20/minimax_h3_ref2v_lightx2v_turbo_4step_v0.1_resized_avg_rank_20_bf16.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9ea3bd3a6aac22994153e294cf1ecab0a8766fc0f8d056ace645a01d1a6a4daf
|
| 3 |
+
size 306731560
|