Update README.md
Browse files
README.md
CHANGED
|
@@ -8,6 +8,13 @@ tags:
|
|
| 8 |
license: apache-2.0
|
| 9 |
---
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
# Boring Embeddings Quick Start
|
| 13 |
|
|
@@ -26,7 +33,14 @@ the embedding's name in your NEGATIVE prompt for more interesting outputs.
|
|
| 26 |
<td style="text-align: center;"><strong style="font-size: larger;">Type the embedding's name (without the .pt extension) in your negative prompt</strong></td>
|
| 27 |
</tr>
|
| 28 |
</table>
|
| 29 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
<br>
|
| 31 |
|
| 32 |
|
|
@@ -39,8 +53,8 @@ depend on manually curated lists of tags describing features people do not want
|
|
| 39 |
* Models might not always understand the tags well due to a dearth of training images labeled with these tags.
|
| 40 |
* It can only capture named concepts. If there exist unnamed yet visually unappealing concepts that just make an image look wrong, but for reasons that cannot be succinctly explained, they will not be captured by a list of tags.
|
| 41 |
|
| 42 |
-
To address these problems, we employ textual inversion on a set of images
|
| 43 |
-
|
| 44 |
hand-labeled artworks which allow users to express their approval of an artwork by either up-voting it or marking it as a favorite.
|
| 45 |
The Boring embeddings were specifically trained on artworks automatically selected from these sites according to the criteria
|
| 46 |
that no user has ever favorited them, and they have 0 or only a very small number of up or down votes. The Boring embeddings
|
|
@@ -49,14 +63,18 @@ the model avoids making mistakes that would make the generation more boring.
|
|
| 49 |
Each training sample consisted of an image paired with a comma-separated list of all the tags associated with it on the site it was sourced from, with the embedding’s name prepended to the list.
|
| 50 |
This ensures that the model learns to associate the embedding with the characteristics of uninteresting images in a way that aligns with the dataset’s tagging system.
|
| 51 |
<br>
|
| 52 |
-
<br>
|
| 53 |
|
| 54 |
|
| 55 |

|
| 56 |
-
<br>
|
| 57 |
-
<br>
|
| 58 |
|
| 59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
## Versions
|
| 61 |
|
| 62 |
### **boring_e621**:
|
|
|
|
| 8 |
license: apache-2.0
|
| 9 |
---
|
| 10 |
|
| 11 |
+
# Overview
|
| 12 |
+
The Boring Embeddings are negative-control textual inversion embeddings for Stable Diffusion models. They are
|
| 13 |
+
widely used in the community (10M+ generations across public tools) to suppress low-quality,
|
| 14 |
+
low-engagement visual patterns and make outputs look more visually appealing.
|
| 15 |
+
<br>
|
| 16 |
+
<br>
|
| 17 |
+
|
| 18 |
|
| 19 |
# Boring Embeddings Quick Start
|
| 20 |
|
|
|
|
| 33 |
<td style="text-align: center;"><strong style="font-size: larger;">Type the embedding's name (without the .pt extension) in your negative prompt</strong></td>
|
| 34 |
</tr>
|
| 35 |
</table>
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
## Research Summary / Technical Contribution
|
| 39 |
+
|
| 40 |
+
This project demonstrates a novel approach to constructing negative-control embeddings using automated quality-based sampling
|
| 41 |
+
from large community-labeled image datasets. The method avoids hand-curated defect lists and instead captures latent features
|
| 42 |
+
correlated with low engagement. Applications include quality control in generative models, aesthetic filtering, and
|
| 43 |
+
domain-specific negative conditioning.
|
| 44 |
<br>
|
| 45 |
|
| 46 |
|
|
|
|
| 53 |
* Models might not always understand the tags well due to a dearth of training images labeled with these tags.
|
| 54 |
* It can only capture named concepts. If there exist unnamed yet visually unappealing concepts that just make an image look wrong, but for reasons that cannot be succinctly explained, they will not be captured by a list of tags.
|
| 55 |
|
| 56 |
+
To address these problems, we employ textual inversion on a set of images extracted from several large community-tagged art
|
| 57 |
+
datasets with rich metadata. Each of these sites is a rich resource of millions of
|
| 58 |
hand-labeled artworks which allow users to express their approval of an artwork by either up-voting it or marking it as a favorite.
|
| 59 |
The Boring embeddings were specifically trained on artworks automatically selected from these sites according to the criteria
|
| 60 |
that no user has ever favorited them, and they have 0 or only a very small number of up or down votes. The Boring embeddings
|
|
|
|
| 63 |
Each training sample consisted of an image paired with a comma-separated list of all the tags associated with it on the site it was sourced from, with the embedding’s name prepended to the list.
|
| 64 |
This ensures that the model learns to associate the embedding with the characteristics of uninteresting images in a way that aligns with the dataset’s tagging system.
|
| 65 |
<br>
|
|
|
|
| 66 |
|
| 67 |
|
| 68 |

|
|
|
|
|
|
|
| 69 |
|
| 70 |
|
| 71 |
+
## Usage & Adoption
|
| 72 |
+
Boring Embeddings have been adopted across multiple Stable Diffusion communities:
|
| 73 |
+
* Downloaded 56k+ times across hosting platforms.
|
| 74 |
+
* Integrated into several popular WebUI / workflow presets as a default negative embedding.
|
| 75 |
+
* Used in 10M+ image generations (based on platform usage statistics).
|
| 76 |
+
<br>
|
| 77 |
+
|
| 78 |
## Versions
|
| 79 |
|
| 80 |
### **boring_e621**:
|