@2929dance Thanks!

ControlNet is king when it comes to really specific stuff. At least all in one go, if you don't want to do inpainting (which can also definitely fix an otherwise perfect output with a quick mask I2I). But stuff like 69 position or crab/lobster insertion, earwax play, etc. a depth map kicks ass. You just pick a nice reference and set a strength. A much more flexible I2I. And you can use it with any model you like with the appropriate apply node. I save depth maps for good outputs that are a pain in the ass to prompt: https://civitai.red/models/1926157/controlnetcreatedepthmap?modelVersionId=2180068

@Ponder_Stibbons Nice thanks! I'm actually really new to all this, local generation and Comfyui stuff, about a month in when I have the spare time. I've watched a video on ControlNet and masking and inpainting so on, just haven't actually gotten around to it...The past few days though I've found the Anime / Cartoon style in Perchance are pretty good in producing the right poses, the real stuff in that sight is always mangled and off though. So what I've done is get a toon/anime picture with the right pose and put it on Flux2Klien Anime to Real slider and it's been pretty awesome.

@Ehh3hh I remember opening comfy for the first time. Vividly. It was just like starting Maya, Blender, After Effects, etc. Feels like you've suddenly got ten thumbs. Even after I had already been using a1111 for awhile, it wasn't exactly pleasant at first. You'll be seeing the birds nest as a clear flow in no time, just keep playing around. Oh since you mention mangled parts here's another rabbit hole for you, albeit a required one imo - use detailer stages if specific body parts are consistently bad- that is just a fact of diffusion. On anything that is not ultra-closeup there is always going to be some feature that is wonky and needs resampling. Face Detailer node isn't just for faces, you just substitute the appropriate bbox/segm detector and it will cut out and fix whatever you need. Chain as many as you need. Lots of detectors around. Collect them like, I don't know, whatever people collect. If you are only using one model (or models with compatible latents) you can pass the latent directly from sampler to sampler. Or use an XL detailer by decoding then encoding for XL. You don't have to drive yourself crazy trying to get a perfect output from the first ksampler. Oh I might as well give you the usual pedantic comfy lecture on backups. A manual backup of the entire comfy folder, leaving out models, input and output, to a dedicated drive will save your hair. Comfy breaks all the time and fixing it can be either five minutes or a whole day. I have many non-fond memories of bricking comfy with one innocuous-looking custom node install before I took five seconds to write a backup script. It's so worth it. Sorry if this is obvious. It's probably obvious. I wish it had been obvious to me.

@Ponder_Stibbons No it's not obvious, at this point I would say I'm a light hobbyist going towards moderate in another half a year so it's actually kind of lot to take in. but I appreciate the input, I'll keep it noted and thanks to youtube I'm sure I can easily look all those up, hopefully. As for the back up stuff, I think that's what happened to my first install within probably 3 workflows that I randomly tried, that was a "normal" desktop install my computer crashed and it wouldn't work again. Now though I'm on a portable install, hasn't bricked yet, so I think maybe I'll back up the whole Comfyui folder even with all the models, and hopefully it will still work even if the folder I'm forking on somehow breaks. thanks for the input.

finally cracked something generally reliable with a mix of loras and i2v

Thanks for the response and the sample. Wish there was a way to steer it away from POV.