improved-aesthetic-predictor VS waifu-diffusion

Perhaps you could use an "aesthetic-evaluator": https://github.com/christophschuhmann/improved-aesthetic-predictor (probably not SoTA but it works)

"Calculates aesthetic score for generated images using CLIP+MLP Aesthetic Score Predictor based on Chad Scorer "

stable-diffusion-v1-2: Resumed from stable-diffusion-v1-1. 515,000 steps at resolution 512x512 on "laion-improved-aesthetics" (a subset of laion2B-en, filtered to images with an original size >= 512x512, estimated aesthetics score > 5.0, and an estimated watermark probability < 0.5. The watermark estimate is from the LAION-5B metadata, the aesthetics score is estimated using an improved aesthetics estimator).

The data used for fine-tuning has come from a random sample of 56k Danbooru images, which were filtered based on CLIP Aesthetic Scoring where only images with an aesthetic score greater than 6.0 were used.

All they've said is they randomly picked 56,000 images that had an aesthetic score greater than 6.0. The score is created by this model. https://github.com/christophschuhmann/improved-aesthetic-predictor

Does anyone know how you could combine a model like Aesthetic Score Predictor with stable diffusion? They used this model to filter training images by score. Iit seems like a lot of people just tune their prompt to make images more aesthetic, by adding certain words.

It's slander to say that the WD team is milking donations in the same way UD is. They already have a public working model which they are continuing to train. The WD team has also implemented features like aspect ratio bucketing into their custom trainer. If they were really milking the project for donations they would have just used the base compvis trainer it was forked from.

Use one of the original SD repos, or the code for Waifu Diffusion, or the Smirkingface refactor.

My model is based off waifu-diffusion.

waifu diffusion (free) and NovelAI (paid afaik) is what I know

This is the one I use: https://github.com/harubaru/waifu-diffusion

This is called textual inversion which use many images and a small group of promtps to describe a single element, and there are two methods: with frozen model and with unfrozen model. The first one just creates a new “word” to “guide” the model, in other words, a word that means the same as a chunk of a prompt. The resulting data is a tiny file. The second one (better known dreamboth training) adds new data to de model, so it can copy the character’s characteristics and artist style more effecively resulting in a new model file (cpkt) There are not a single tutorial, so you can find many info on r/StableDiffusion. Also, there is a third method which is based on the original training (many images and its own description) Instruction are here: Training guide I don’t know if this guide is finished though

The model is a tweaked version of waifu-diffusion using textual inversion. Still not perfect but the AI can create pretty plausible images.