Introverted_69

So, because it's such a big undertaking, I thought I'd let people know what I'm up to. I'm building what should be an about 450 ish picture hand processed dataset of background and people from: old west, 19th century Victorian, fantasy/medieval, post-apocalypse and cyberpunk. These are mostly photographs/stills, that have been processed and cleaned up, with some high quality renders and AI renders. I still have probably about a month of work to go on that. If it works, as a DORA, it should be able to do period accurate lighting, windows, more realistic materials and people that are less in an illustration style. Cyberpunk should be more realistic, understated and believable too. In particular, I'm hoping stone, concrete, wood textures come out a lot better for flux, candle holders, banquet lanterns, lattice windows, wooden internal shutters, period accurate dress, makeup and hairstyles. Hopefully no more cartoon stones, high heels, eyeshadow, electric lighting and modern windows where they are not wanted. After that, the next big project is a POV adapter, not as a standalone POV lora/dora but so that POV lora's can interact with it, and the base model whilst retaining genre/period accuracy. This way if you use some POV lora, you hopefully won't suddenly get anachronisms. In the meantime, I've injected a very small amount of hairstyle's looks into the upcoming version of Elegant Nudes, that can act as an SDXL companion to the larger project. I'll post that when I can get it trained right (dataset is complete). That expands the initial dataset for elegant nudes some, will be trained instead on a big love base for hopefully more realism and flexibility, and now includes roughly 30% or so genre/period pictures, in addition to the general vibe of artful/tasteful nudes.

Lora training lessons learnt as a newbie trainer.All this applies to realistic pictures and may apply less so to highly stylized or illustrated loras. 1) One or a few pictures in the dataset can contaminate the whole set. Anything that leans overly in a direction, such as bright lighting, bright colors or the opposite can be picked up on and ran with. Stylistic dataset variety even in photos is ideal. Recommend about 10-20 pictures with slight different style or lighting conditions. 2) Autotagging is next to useless. Good tagging makes all the difference for manual control of outputs, so either manually carefully edit or create every tag set, or don't even both tagging. Bad tags can ruin a train. This can also allow you to properly spot jewelry, tattoos, watermarking, logos and the like which AI cannot do well, and also allow you to add in variables the end user can control such as ethnicity, eye and hair color, body type, situations or body positions, and context like where the background is. If you do this poorly, it results in poor control - but as I said, poor tags can also ruin the train. If niche features, like say bright studio lighting, or semirealistic are funneled into tags they are more limited in input and also can be negatively prompted for. 3) 50-100 pictures is the ideal size. 100 in particular allows a few examples of multiple subthemes if they are properly tagged. You are better to manually curate these, and manually crop out the logos/text if possible. That will give you better picture quality and diversity than just scrapping. 4) As with bright pictures, or dark pictures, synthetic data can spoil the batch. I find no more than 40% synthetic data, focusing only on high quality data like the best of Pony XL, or best of Flux to be ideal. You need to be choosy when it comes to synthetic data. 5) Slow training allows for more generalization.6) Higher resolutions generalize more than lower resolutions. This can also be a bad thing if your dataset, or tags spoil the batch. If your tagging or dataset is not ideal, you might go for 512x512, although the results will be worse on details.7) Increasing the batch size vastly increases the generalization, which can be a good or a bad thing depending on the subject material. I'd say outside of a very narrow use, like say a body posture, or clothing item, this is probably not ideal, as it can overgeneralize strongly. I probably have a lot more to learn here, but these are a few things I've picked up.