Tags
Not in edit mode!
(This will go through all models and add tags if necessary to reflect current tag implications.)
Subject
Images or videos rendered with tools such as Stable Diffusion
Image or video of anime or anime-style artwork.
Image or video of western cartoon or artwork in a similar style.
Computer Generated Imagery (CGI). These are images or videos that have been rendered in software such as Blender. A good example would be the movie "Toy Story"
Image or video of faces. This can be photograph or artwork.
Images (screenshots) taken of video games.
Textures used in video games or other 3D applications. These textures are typically diffuse/albedo maps, but other types of textures (such as normal maps) are also included.
Image of manga pages or manga-style artwork.
Pixel art. This includes things such as retro pixel artwork (think SNES games).
Image or video of realistic looking imagery. E.g. images recorded with camera.
Image or video of text.
Images of individual video frames.
Purpose
The model smooths aliasing artifacts (also called jaggies).
The model colorizes grayscale and/or black-and-white images.
The model is capable of removing compression. This can be any type of compression, such as JPEG or H264.
The model is able to remove DDS block compression artifacts.
Models with this tag may only support specific compression methods and/or settings (e.g. BC1/DTX1, BC7).
The removes banding artifacts caused by color quantization. Such artifacts and frequently be observed in GIF images.
The model tries to remove blur from the input.
The model removes dithering artifacts from the input.
The model removes halftone patterns from the input. This is typically on printed media, such as magazines or album covers.
The model removes haloing and color bleeding from the input.
The model removes noise from the input.
Models with this tag may only work with specific kinds of noise.
The model can upscale any subject.
General purpose models achieve good results across a wide range of inputs. However, models that specialize in the subject of a specific image (e.g. anime, faces, text, etc.) will typically achieve better results than general purpose models.
The model fills in selected parts of an image.
The model is able to remove JPEG compression artifacts.
Models with this tag may only support specific JPEG compression settings (e.g. quality between 80% and 100%).
The model is intended to be used to train new models.
The model is the result of a research paper. The paper or GitHub repo is typically linked in the description.
The model restores images or video frames through compression removal, artifact clean up, or otherwise.
The model generates textures from other textures. E.g. a model could generate a normal map from a diffuse texture.
ArchitectureEdit Architectures
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
No description
Platform
The model has PyTorch models (.pth/.pt) to download.
The either has PyTorch models to download or can be converted into a PyTorch model.
The model has ONNX models (.onnx) to download.
The either has ONNX models to download or can be converted into an ONNX model.
The model has NCNN models (.bin+.param) to download.
The either has NCNN models to download or can be converted into an NCNN model.
Scale
No description
No description
No description
No description
No description
No description
Input type
No description
No description
No description
Color
The model upscales grayscale images. A typical example for grayscale models are models that upscale the alpha channel of images.
The model upscales images without transparency. Some applications (e.g. chaiNNer and CupScale) can still upscale images with transparency using this model.
The model upscales images with transparency.
The model colorizes grayscale images to RGB.
License
Crediting the author is required when sharing the model or derivative works (e.g. model interpolations).
Simply using the model to upscale images is typically okay, and no credit is required for upscaled images.
The model may be included and used in commercial products.
The author allows that this model may be used to create derivative works (e.g. model interpolations).
The author requires that derivative works have the same license or a compatible license as the model. Derivative works include model interpolations and using it a pretrained model.
The model either has no license or is licensed under a custom or infrequently used license.
Helpers
The model has no description.
The model has no tags.
The model has no date or its date is invalid.
The scale of the model is not valid.
The number of input or output channels of the model is not valid.
Dataset
Realistic dataset
Anime dataset
Manga dataset
Game textures dataset