Type | |
Stats | 3,327 |
Reviews | |
Published | Aug 14, 2023 |
Base Model | |
Trigger Words | ageprogression |
Hash | AutoV2 9D622AC348 |
Image posts are highly appreciated! You don't even need to leave a review. I just love to see what people are creating with my stuff.
Produces an image of a single character at several ages/growth stages/body styles at a left to right slant. The triggerword is ageprogression. Can be used with terms such as full body, cowboy shot, and upper body to adjust the view of the presented character.
Cowboy shot and upper body generally give 2-3 bodies per image.
Full body generally gives 3-4 bodies per image, occasionally 5.
You can somewhat direct the age level by using keywords related to breast size tags, (flat chest, small-medium-large-huge-gigantic breasts) or ages (adult*, mature female, other...)
*Not an actual Danbooru tag but any age descriptors seem to work fairly well.
Wider images will either allow for more bodies to fit in or will just give more breathing space for what's there. As a person limited to 8gb of vram, I'm able to get 640x512 images with 2x upscale, and 700x500 images with 1.8x upscale while using the DPM++ SDE Karras sampler and the R-ESRGAN 4x+ Anime6B upscaler. These sizes should work unless your sampler/upscaler use a higher amount of vram than mine.
Had successful and good/decent quality image outputs on Hassaku-v1.3, Koji-v2.1, Anything v4.5-pruned, and the old NAI leak. Had fairly bad outputs when attempting testing on AbyssOrangeMix3-A1B.
There's currently no way to specifically set the amount of bodies that appear per image. Mainly due to the difficulty in training separate body counts given that (a) there's not much age progression art out there, and what is out there is mostly low quality sketches or age charts, and (b) that out of the few applicable images that I was able to grab, a majority of them are 2/3 bodies per image so any keywords for 4+ characters would likely be iffy in their consistency. I might later attempt to manually create a data set of 4, 5, 6, character images by generating single "characters" at different stages and photoshopping them together, but I've got this fear that attempting to create a LoRA out of AI incest would have mixed results.
Also, it runs into an issue with character LoRAs. Most age progression images have characters at various ages and not wearing the same clothing, which when mixed with character LoRAs has the effect of making the characters' trained outfits somewhat consistent.