Natural language works (wearing diapers, wearing a diaper) and all diapers generated will be from the selected brand.

Only two tapes will likely be generated when the available space is smaller, such as when there are multiple people or the subject is far away.

Pacifiers in mouths were included in the training data and can be prompted.

Available to Train: (no particular order)

• ABU BunnyHops Four Tape Blue

• ABU DinoRawrZ (alternative dino)

• ABU LittlePawz

• ABU PeekABU (Available from Another Creator, See Suggested Resources)

• ABU TinyTails

Prior Development Notes

I really thought having some examples of actual two tape bunny hops would have fixed the tape issues, but no. Even overtraining a bit didn't increase or decrease the likelihood of a four or two tape image. Part of me wants to blame the distilled model, but until we get a decent fine-tune of SD3.5L/M, it's just speculation.

That said, I feel good about my hook and loop tape diaper dataset recipe even if the number of tapes isn't perfect in every generation.

DinoRawrZ major defect beyond the expected smushed two tapes issue is that sometimes the front print will simply not appear. I picked an epoch where it happened the least. Might be a landing strip associated issue that I just didn't notice with BunnyHops.

AlphaGatorZ tapes sometimes blend into the diaper. BabyParade has it worse though. I think it's the same root cause issue that is making sticky tapes not work out for me.

A little ink around the edges of the tapes for a few shots did wonders for Little Kings. Now I just need to figure out an equivalent trick for clear sticky tapes.