Update README.md
Browse files
README.md
CHANGED
|
@@ -111,13 +111,13 @@ You may be interested in comparing Anima's outputs with other models. A ComfyUI
|
|
| 111 |
- The preview model is a true base model. It hasn't been aesthetic tuned on a curated dataset. The default style is very plain and neutral, which is especially apparent if you don't use artist or quality tags.
|
| 112 |
|
| 113 |
# Finetuning Tips
|
| 114 |
-
- Any LoRA you train on a preview version should be considered a "throwaway" LoRA. There's no guarantee it will work well on the final version.
|
| 115 |
- **Don't train the LLM adapter.** My own training script, diffusion-pipe, lets you set llm_adapter_lr=0 to completely disable training it, and the example config has this as a default.
|
| 116 |
- Other trainers like sd-scripts have similar options that should be used.
|
| 117 |
- The LLM adapter processes the text embeddings before they get to the diffusion model, and therefore has an outsized influence on the generated images. The adapter itself contains a surprising amount of knowledge and is easy to degrade by training it.
|
| 118 |
- Use a low learning rate. For a rank 32 LoRA, start with 2e-5 and adjust up or down from there.
|
| 119 |
- As a base model, there is no aggressive aesthetic tuning or RLHF you need to overcome when finetuning.
|
| 120 |
- The model has an extremely large and diverse amount of visual concepts baked in already. A light touch is all you need.
|
|
|
|
| 121 |
|
| 122 |
# License
|
| 123 |
This model is licensed under the CircleStone Labs Non-Commercial License. The model and derivatives are only usable for non-commercial purposes. Additionally, this model constitutes a "Derivative Model" of Cosmos-Predict2-2B-Text2Image, and therefore is subject to the [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/) insofar as it applies to Derivative Models.
|
|
|
|
| 111 |
- The preview model is a true base model. It hasn't been aesthetic tuned on a curated dataset. The default style is very plain and neutral, which is especially apparent if you don't use artist or quality tags.
|
| 112 |
|
| 113 |
# Finetuning Tips
|
|
|
|
| 114 |
- **Don't train the LLM adapter.** My own training script, diffusion-pipe, lets you set llm_adapter_lr=0 to completely disable training it, and the example config has this as a default.
|
| 115 |
- Other trainers like sd-scripts have similar options that should be used.
|
| 116 |
- The LLM adapter processes the text embeddings before they get to the diffusion model, and therefore has an outsized influence on the generated images. The adapter itself contains a surprising amount of knowledge and is easy to degrade by training it.
|
| 117 |
- Use a low learning rate. For a rank 32 LoRA, start with 2e-5 and adjust up or down from there.
|
| 118 |
- As a base model, there is no aggressive aesthetic tuning or RLHF you need to overcome when finetuning.
|
| 119 |
- The model has an extremely large and diverse amount of visual concepts baked in already. A light touch is all you need.
|
| 120 |
+
- [Example](https://civitai.com/models/2536147) of a style LoRA, with dataset and configs shared.
|
| 121 |
|
| 122 |
# License
|
| 123 |
This model is licensed under the CircleStone Labs Non-Commercial License. The model and derivatives are only usable for non-commercial purposes. Additionally, this model constitutes a "Derivative Model" of Cosmos-Predict2-2B-Text2Image, and therefore is subject to the [NVIDIA Open Model License Agreement](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/) insofar as it applies to Derivative Models.
|