CharTurner

Not better, just different? I think Lora are a little heavy handed (for instance, the Lora version has to be used at .2 - .4 weighting, or it takes over entirely.). I'm new to Lora, so i probably bollixed up some setting, hope to have it worked out and behaving soon.

@juvesatriani StollenDreams, Dreamshaper, RealisticVision, 1.5 pruned (which is what it was trained in). Maybe a few others here and there but those are my main ones.

Oh and Protogen58scifi

Maybe take a 25 pose shot of a different character and try img2imge over it? With this version, it's probably not doable to prompt for 25 specificic poses.

22 images, 660 steps, about an hour to train. BUT i've been working for WEEKS, training over and over, trying to find the right balance of dataset and settings. There's a bug in auto1111 training with xformers that might be screwing you, check this out (link should go to workaround solution that fixed it for me) https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/7264#issuecomment-1418206085

Also, I wrote a thing for expression sheets, because I had no luck with an embed either, really. This method works tho https://www.reddit.com/r/StableDiffusion/comments/103rk2k/use_script_xy_prompt_to_create_expression_sheets/?utm_source=share&utm_medium=web2x&context=3

Yup, that's one way to do it. LIke, get a set of dummies that are in roughly the right spots, put your character over ONE of them, mask that one, and use an inpainting model to img2img the rest of them.

various, including base 1.5 pruned, stollenDreams, Deliberate, Dreamshaper, Protogen58, RealsticVision.

I can't spell. XD https://civitai.com/models/2540/elldreths-stolendreams-mix

anything wider than it is tall, i don't have a ideal dimension. All the source images are square, of course, but the many turn ones have a lot of 'headroom', which the ai mostly doesn't like. So, i've tried everything from 4:3 to 16:9 and they all work. (there's an extension that adds ratio buttons to automatic1111 now, they are very useful). So, 512 for the height, something larger for the width. 768 works well just cuz it's tidy. XD

Steps depend on your sampler. I do a lot of ddim at 20-30 steps.

you can try! :D I didn't teach it to do that specifically but you can probably get it to happen.

I don't think there's a safetensor version of textual embeddings, but if I can find out how to do that I will. :D

Checked with the Civatai people; no way currently to have a safetensors embed. There IS a lora version, though it's not as good yet.

I don't think that gui takes textual inverserion embeds, sadly. I will ask around, but afaik, it doesn't.

I have it ready to go, just trying to get a few more demo shots, so probably today?

Ooo, I dunno, that'd be rad if so

Update: yes, yes it can. Control net will pick up multiple people, and using charturner means they all get the same outfit. :D

Nope. It's not a model, and there's no safetensors verison of embeds (if they make one, I will use it, but it doesn't currently exist). This goes on top of your model.