Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,35 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model:
|
| 4 |
+
- MiniMaxAI/MiniMax-H3
|
| 5 |
+
- lightx2v/Minimax-h3-Turbo
|
| 6 |
+
base_model_relation: adapter
|
| 7 |
+
pipeline_tag: text-to-video
|
| 8 |
+
tags:
|
| 9 |
+
- lora
|
| 10 |
+
- t2v
|
| 11 |
+
- i2v
|
| 12 |
+
- comfyui
|
| 13 |
+
- minimax-h3
|
| 14 |
+
---
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
## Contents
|
| 18 |
+
|
| 19 |
+
| file | size |
|
| 20 |
+
|---|---|
|
| 21 |
+
| `loras/minimax_h3_fl2v_lightx2v_turbo_8step_merge_0821_bf16.safetensors` | 1.96 GB |
|
| 22 |
+
|
| 23 |
+
## Recipe
|
| 24 |
+
|
| 25 |
+
### FLFV 8 Step Merge 0821
|
| 26 |
+
|
| 27 |
+
| source | weight |
|
| 28 |
+
|---|---|
|
| 29 |
+
| `minimax_h3_fl2v_turbo_4step_v1.1_768p_comfyui_bf16.safetensors` | 0.5 |
|
| 30 |
+
| `minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors` | 1.0 |
|
| 31 |
+
|
| 32 |
+
The 4 step LoRA was trained using shift 6 and the 8 step was trained using shift 12,
|
| 33 |
+
so theoretically you can't go wrong with any shift in that range. Tested at strength
|
| 34 |
+
1.0 and shift 9, which gives more detail than v1.1 LoRA and fixes poor fingers
|
| 35 |
+
rendering on the 8 step 1.0 LoRA.
|