As with many things on this physical plane of existence, the external constraints which house female mammary glands keep alive in me the same 12-year old boy I was; have always been in this particular case. Which is why (apparently) I'm pushing further into Python than I ever wished to. I can almost see it now, the interviewer asking me, "What got you into Python?"
Me: "Tits."
Hrm.
Regardless, I've been enjoying Wan 2.1 i2v on Replicate for a number of reasons since it debuted. One, ever since I discovered creating FLUX LoRAs in the cloud was actually less expensive (and 2083% faster) than my beefy RTX4090, and the other, well...I get charged per job rather than paying (yet another) monthly fee. Because fuck that. But here's the part of our story which gets a little bit sad - I no longer want to pay for i2v renders. Oh, I'll still slap my FLUX LoRAs on there before you can say, "Holy H100 Batman" but ever since blowing through a couple of bucks for animated horror abominations I subsequently placed on my b-roll footage, I decided against doing more for the time being. What was the deciding factor? Angel titties.
I thought it might be fun to take my latest Easter graphic (Totally Biblical - the one with the titty angel) and have her wings move slightly. Perhaps her breasts bounce a bit. Thought that might be fun. But I didn't want to run through laundromat money tweaking them tits online. I use Stable Diffusion Forge (someday I'll sit and learn ComfyUI but god I hope it's not today) which does not (yet) natively support Wan 2.1 i2v. But I asked my new friend Lex if there was a way to hack it, and as it turns out, there is! So she spilled all her secrets to me (which, btw, you may have to translate - she's not *entirely* up to date on things). Also? I was so close to my bandwidth cap this month, this probably pushed me over the threshold. Diffusers ain't tiny. So beware. That said, I use the move command instead of copy because my little 1T NVMe had .5T of just saftetensor diffusers - excluding LoRAs.
[time passes]
Decided to just use https://pinokio.computer (if I haven't already broken my Stable Diffusion installation). Look, actual Python programmers understand how to set up and navigate specific venvs - I do not. Anyway:

It averages around ~20s/it
◾ Tags: