Document model training limitations
Browse files
README.md
CHANGED
|
@@ -18,6 +18,15 @@ tags:
|
|
| 18 |
|
| 19 |
# LPC Four-Direction Walk — FLUX.2 Klein 9B
|
| 20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
An experimental rank-16 LoRA for editing female LPC characters and 4×2
|
| 22 |
walking sprite sheets with **FLUX.2 [klein] Base 9B**.
|
| 23 |
|
|
@@ -28,6 +37,10 @@ walking sprite sheets with **FLUX.2 [klein] Base 9B**.
|
|
| 28 |
These GIFs are direct GUI exports. Samples 1, 2, and 5 contain 24 frames;
|
| 29 |
samples 3 and 4 contain one 8-frame walking cycle.
|
| 30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
| Sample 1 | Sample 2 | Sample 3 |
|
| 32 |
|---|---|---|
|
| 33 |
|  |  |  |
|
|
@@ -85,6 +98,9 @@ TASK_DRESS_STANDING: Dress the south-facing standing female LPC character with b
|
|
| 85 |
## ComfyUI workflow
|
| 86 |
|
| 87 |
The workflow uses only ComfyUI core nodes and has been tested on Comfy Cloud.
|
|
|
|
|
|
|
|
|
|
| 88 |
|
| 89 |
1. Install
|
| 90 |
[`LPC_FourDirection_Walk_ComfyUI.safetensors`](https://huggingface.co/Baragi-AI/LPC-FourDirection-Walk-Flux-Klein-9B/resolve/main/LPC_FourDirection_Walk_ComfyUI.safetensors)
|
|
|
|
| 18 |
|
| 19 |
# LPC Four-Direction Walk — FLUX.2 Klein 9B
|
| 20 |
|
| 21 |
+
> **Warning**
|
| 22 |
+
>
|
| 23 |
+
> This experimental LoRA was trained on very little data, without adequate
|
| 24 |
+
> regularization, and at a rank that is low for the number and complexity of
|
| 25 |
+
> its tasks. Training a new LoRA for your specific goal is recommended.
|
| 26 |
+
> An image-edit-capable base model should produce better results when trained
|
| 27 |
+
> on fixed-integer-upscaled pixel art with consistent trigger words and a
|
| 28 |
+
> deliberate mixture of text-to-image and image-edit examples.
|
| 29 |
+
|
| 30 |
An experimental rank-16 LoRA for editing female LPC characters and 4×2
|
| 31 |
walking sprite sheets with **FLUX.2 [klein] Base 9B**.
|
| 32 |
|
|
|
|
| 37 |
These GIFs are direct GUI exports. Samples 1, 2, and 5 contain 24 frames;
|
| 38 |
samples 3 and 4 contain one 8-frame walking cycle.
|
| 39 |
|
| 40 |
+
Outputs may be inconsistent, as samples 2–4 demonstrate. Poorly written
|
| 41 |
+
training captions made this LoRA highly prompt-sensitive, so small prompt
|
| 42 |
+
changes can produce results that are difficult to predict.
|
| 43 |
+
|
| 44 |
| Sample 1 | Sample 2 | Sample 3 |
|
| 45 |
|---|---|---|
|
| 46 |
|  |  |  |
|
|
|
|
| 98 |
## ComfyUI workflow
|
| 99 |
|
| 100 |
The workflow uses only ComfyUI core nodes and has been tested on Comfy Cloud.
|
| 101 |
+
Because the LoRA is undertrained, using it with PAG is recommended. A LoRA
|
| 102 |
+
strength of `1.2` can produce better results than `1.0` for the same reason;
|
| 103 |
+
these values are starting points rather than guarantees.
|
| 104 |
|
| 105 |
1. Install
|
| 106 |
[`LPC_FourDirection_Walk_ComfyUI.safetensors`](https://huggingface.co/Baragi-AI/LPC-FourDirection-Walk-Flux-Klein-9B/resolve/main/LPC_FourDirection_Walk_ComfyUI.safetensors)
|