Epic anime landscape4/16/2023 ![]() ![]() This model could also work on some real life images, especially the ones that are taken outdoors. From my testing, this works rather well on realistic GameCube textures such as the ones from Shrek Extra Large and the board textures from Mario Party 4. This model was made to upscale realistic low-res textures that are compressed by either JPEG or BC1. Universal upscaler for clean and slightly compressed images (JPEG quality 75 or better) Trained using the patchgan discriminator, with cx loss, cutmixup and frequency separation, it produces good results with a slight grain due to patchgan, with some sharpening using cutmixup. Colors are pretty good as well as edges, but generated details seem slightly fuzzy hence the name.Ī universal model, that is aimed at prerendered images, but handles realistic faces, manga, pixel art and dedithering as well. Photographs, Artwork, Textures, Anything really - Tried out a new pixel loss idea based on ensuring the HR downscaled matches the LR. NOTE: THIS WILL NOT WORK IN CUPSCALE, IEU, OR CHAINNER (yet)! You have to use my fork to use it for now. Basically it makes smaller ESRGAN models that theoretically can produce the same level of quality. You can read more about it in the github README. Description: Pretrained model for the new architecture modification I made. Technically my previous experiment was the pretrained model, but for all intents and purposes this was trained from scratch. Supposed to help retain more details, but unfortunately due to the dataset (I think) still blurs details adjacent to other objects. DO NOT USE FOR COMPRESSED IMAGES, use the original UniScale or UltraSharp for that.īasically realesrgan-x4plus without the degradation training. These models work great on game textures when interpolated 50/50 with UniScale_Restore, and work amazingly on uncompressed images. Trained with BSRGAN_Resize and Combo_Noise in traiNNer. UniScale_Restore has strong compression removal that helps with restoring heavily compressed or noisy images. This model removes noise from images while upscaling. Version of UniScale trained with camera noise injection (NR = Noise Removal). It was originally intended to upscale game textures, but was expanded into a universal upscaler. This model can upscale almost anything well. UniScale strikes a nice balance between sharpness and realism. ![]() It's a bunch of interpolated models based around UltraSharp and my other models It has the ability to restore highly compressed images as well! If you want a more balanced output, check out the UltraMix Collection down below. It does work best on JPEG compression though, as that's mostly what it was trained on. It works on most images, whether compressed or not. This is my best model yet! It generates lots and lots of detail and leaves a nice texture on images. This was, things like skin and other details don't become mushy and blurry. Outdoor scenes.Ī creation of BSRGAN with more details and less smoothing, made by interpolating IRL models such as Siax, Superscale, Superscale Artisoft, Pixel Perfect, etc. Streets with dense foliage in the background. Image scaling and Video upscaling Universal Models Model Name ![]() These programs can be used to train your models: BasicSR, the official ESRGAN repository (old arch tag), victorca's traiNNer ( ), or sudo's colab-traiNNer ( ). The others are: IEU by Honh Cupscale by NMKD. The only actively maintained program is chaiNNer by Joey. There are various GUIs available to inference/upscale with these models. These are all models that use the "old" ESRGAN architecture. 1.2.12.2 Normal Map/Bump Map Generation.1.1.4.1.1 Anime and Cartoon Restoration. ![]()
0 Comments
Leave a Reply.AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |