Update README.md

README.md

@@ -18,8 +18,7 @@ language:
 
 FluffyRock Unbound is a finetune of [Fluffyrock Unleashed v1.0](https://huggingface.co/RedRocket/Fluffyrock-Unleashed) trained on an expanded, curated <a href="https://e621.net/">e621</a> dataset and with training changes adapted from Nvidia Labs <a href="https://arxiv.org/abs/2312.02696">EDM2</a>.
 
-This model can produce detailed sexually explicit content and is not suitable for use by minors.
-It will generally not produce sexually explicit content at all unless prompted.
+This model can produce detailed sexually explicit content and is not suitable for use by minors. It will generally not produce sexually explicit content unless prompted.
 
 ## Downloads
 
@@ -27,16 +26,7 @@ It will generally not produce sexually explicit content at all unless prompted.
 <a href="https://huggingface.co/RedRocket/Fluffyrock-Unbound/resolve/main/Fluffyrock-Unbound-v1-1.yaml?download=true">Fluffyrock-Unbound-v1-1.yaml</a> - YAML file for A1111 Stable Diffusion WebUI. Place this in the same folder as the model.<br>
 <a href="https://huggingface.co/RedRocket/Fluffyrock-Unbound/resolve/main/fluffyrock-unbound-tag-strength-v1.1.csv?download=true">fluffyrock-unbound-tag-strength-v1.1.csv</a> - Recommended tag completion file, representing the strength of each concept in the model. (<a href="fluffyrock-unbound-tag-strength-v1.1.csv
 ">Raw Counts</a>, <a href="https://huggingface.co/RedRocket/Fluffyrock-Unbound/resolve/main/fluffyrock-unbound-tag-metadata-v1.1.csv?download=true">Metadata</a>)<br>
-
-### TI Embeddings
-
-<a href="https://huggingface.co/RedRocket/Fluffyrock-Unbound/resolve/main/boring_e621_unbound_lite.safetensors?download=true">boring_e621_unbound_lite</a>
-<a href="https://huggingface.co/RedRocket/Fluffyrock-Unbound/resolve/main/boring_e621_unbound_plus.safetensors?download=true">boring_e621_unbound_plus</a>
-
-Currently experimental textual inversion embeddings in a similar vein to the [Boring Embeddings](https://huggingface.co/FoodDesert/Boring_Embeddings).  Intended to improve quality while not drastically altering image content.  Intended to be used as part of a negative prompt (although using them in the positive prompt will work and may be even more fun).
-
-The "lite" version is 6 tokens wide and is initialized on the values of ``by <|endoftext|><|endoftext|><|endoftext|><|endoftext|><|endoftext|>``, which is very close to a "blank slate".
-The "plus" version is trained on the same dataset, is 8 tokens wide, and is initialized on an average vector of 100 low-scoring artists.  Currently, the "lite" version is recommended.
+<a href="https://huggingface.co/RedRocket/Fluffyrock-Unbound/resolve/main/boring_e621_unbound_lite.safetensors?download=true">boring_e621_unbound_lite.safetensors</a> - Boring-E621 style negative embeddings to improve quality. Use in the negative prompt. (<a href="https://huggingface.co/RedRocket/Fluffyrock-Unbound/resolve/main/boring_e621_unbound_plus.safetensors?download=true">Stronger Plus Version</a>)
 
 ## Prompting Guide
 
@@ -57,6 +47,14 @@ Place the model checkpoint in the ``models/checkpoints`` folder. The optional Bo
 The model is zero-terminal-SNR with V-prediction. Use the ModelSamplingDiscrete node to configure it properly.
 <img style="margin-top: 0; width: 500px;" src="https://huggingface.co/RedRocket/Fluffyrock-Unbound/resolve/main/comfyui-1.webp">
 
+## Quality Embeddings
+
+Experimental textual inversion embeddings in a similar vein to the [Boring Embeddings](https://huggingface.co/FoodDesert/Boring_Embeddings) are provided above.
+They're intended to improve quality while not drastically altering image content. They should be used as part of a negative prompt, although using them in the positive prompt can be fun too.
+
+- The "lite" version is 6 tokens wide and is initialized on the values of ``by <|endoftext|><|endoftext|><|endoftext|><|endoftext|><|endoftext|>``, which is very close to a "blank slate".
+- The "plus" version is trained on the same dataset, is 8 tokens wide, and is initialized on an average vector of 100 low-scoring artists. Currently, the "lite" version is recommended.
+
 ## Training Details
 - Adaptive timestep weighting: Timesteps are weighted using a similar method to what the EDM2 paper used, according to the homoscedastic uncertainty of MSE loss on each timestep, thereby equalizing the contribution of each timestep.  Loss weight was also conditioned on resolution in order to equalize the contribution of each resolution group.  The overall effect of this is that the model is now very good at both high- and low-frequency details, and is not as biased towards blurry backgrounds.
 - EMA weights were assembled post-hoc using the method described in the EDM2 paper.  The checkpoint shipped uses an EMA length sigma of 0.225.