Fusing custom trained checkpoint or lora
Fis solution
Fwhen we train our fine tuned models it also remove that
Ftrue
MI'm using one
MBut it's true that maybe It's not the perfect dataset
MYep I see ... Any other model recomendations to get realistic portraits?
Tyou can't get better results than flux, you just need more effort with it. Consider better prompt engineering
AGitHub
Contribute to ali-vilab/ACE_plus development by creating an account on GitHub.
Tnope
Aany tips on better prompt engineering? is there like a custom gpt out there for flux prompting ?
Tnope, just experiment. There is no single solution that fits all cases
TAlso img2img is a powerful tool, even if you can't get result just with prompt you can use reference images and/or inpainting
Dtheres a myriad of options available. filmic looking loras, checkpoints trained for analog look, you don't need to limit yourself for base flux
FF
SGuys, is it worth upgrading from rtx3060 12gb to rtx4060ti 16gb for local training?
Fit will be a small improvement
F4070 super 16 gb better
S4070 super 16gb same price as 3090 24gb :/
SSo probably 3090 better of those 2?
Fin that case 3090 better :d
Fi recommend used 3090
SProbably will have to go this route. Thanks 
Fi moved my home setup to work and purchased 2nd used 3090 for it :d
Fdmn isnt this impressive
Ahow do I use image2image in swarm? I haven't played with it
Tit's all in the Dr's video tutorials
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dnah no seal noise 
Dso is this actually an updated improved model for mmaudio doc?
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fi am adding more features right now
F
CI love it.
CAmazing. Can you share prompt and settings?
CThe physics of water, etc. Beautiful.
Fi will show in tutorial how to prompt
Fespecially video and images are easy to generate with ai
DAh yeh it does images now
Vdoc, in kohya lora training do you know if there is a difference between using 240mb clip L vs 1.6gb clip L. Is this something you have tested?
Fyes i tested
Fi feel like 1.6gb clip a bit better
Dthe 1.6gb clip L is larger
D
Dhunyuan img to video is dissapointing ive heard
VYea, im just trying to figure out if that's something I can trade off on to cut training time. On 4090 tier_4 (lowest of the 4 configs) on 200 epochs im getting a 5.5hr ETA on training for 30 images for FLUX. Intermediary checkpoints are cooking well, but the time it wants to take is a bit of a bummer.
Dcloud services ?
Vno local
V2hrs felt normal for sdxl/sd15 training. Just getting into it on the flux side. Not sure if this is what the average is across the board.
Drent a more powerful gpu to cut the time
