I normally post things in the AI WAM group but wanted to drop a bit of a larger sampler here of recent stuff and a few upcoming ones. The quality is definitely improving with what's possible. I train my own AI models to produce these images and only train using real WAM images I shot myself.
I know there is always interesting debate with AI content but the nice thing about a community is that everyone can like different things
I post all my sets (both realistic and anime) in my deviantart page which is linked from http://messy-ai.com . The aim with the site ultimately is to be able to create a tool in there that allows anyone to generate their own WAM images. That's a bit off yet but hopefully on the cards.
Hope you enjoy them and always happy for feedback.
**Disclaimer for compliance with UMD rules:**
1. All training for my messy AI models are done using photos which I produced and own copyright and usage rights for. No material belonging to anyone else was used in the training process. 2. Models have been trained for style and concept keywords only and not to replicate any real persons image or likeness. 3. All generated images depict subjects above the age of 18
Looking really good! Just a tip for your captions: Make sure to include the things that you want control over. In your case mentioning: "pouring from above"/"dripping from above", "asian", and "arms held up", could help to provide some more variety. I'd highly recommend taggui, which lets you use local LLM models to help automate portions of the captioning process.
ABGamma said: Looking really good! Just a tip for your captions: Make sure to include the things that you want control over. In your case mentioning: "pouring from above"/"dripping from above", "asian", and "arms held up", could help to provide some more variety. I'd highly recommend taggui, which lets you use local LLM models to help automate portions of the captioning process.
Thanks. Yeah I've been using moondream and cogVLM in the locally run GPT4V image captioner. They are a solid start which I then adjust the caption manually to use the more precise wording I want. Similarity in style is more driven by my using only WAM images to train on that I shot myself, which definitely have a style to them. I think that is important though as it would be super easy to train on another producers hard work which I don't think is right without permission in a small community like this.