Can't wait to download your config with SD 3.5 and test and pay 5 usd more
Can't wait to download your config with SD 3.5 and test and pay 5 usd more
Fsure you can
Fso let me clarify
Fwith my config
F100 epoch trained
Fsaved
Fthen continued another 100
Fshould be equal to directly training 200
Wthank Doc
Wto continue training I need to tick on these option right ? Maybe you can make a guide about how to continue train a model which require very long training time by separate it into many training section
Fyou can use saved checkpoint
Fthat is not necessary
WFor example I trained 200 epoch but stop at 100 epoch and saved some checkpoint like this
W
WDo I need to recalculate the training rate
Wor I just need to change the epoch to 100 epoch because I already train 100 epoch
Fjust change epoch
Fthats it
Wthank you so much
B
Frtx 4090 will be probably faster with newest fast branch
BHas the newest branch been updated to the script?
Fdoes some one have any tips on workflows for comfyui to play around with my finetuned FLux Dev safetensor? Maybe some Lora to make it faster?
F@Furkan Gözükara SECourses is there a way to make it faster like schnell or fp8 after I trained a full fluc dev model?
Ffurther more , any face detailers workflows for comfyui using yolo like you have on Swarm?
Fcheers
FGitHub
Contribute to kohya-ss/sd-scripts development by creating an account on GitHub.
Fi think no way
Fwhat gpu you have?
Frtx4090
Fok then it should be pretty fast already
Fwhat is your step speed?
Fis fast but not as fast as shcnell right?
Fyep
Fschnell fast comes from lesser step
Fbut step speed should be same
F
Fis this flux?
Fso slow
Fuse swarmui
Fit is 2.2 it / second
Fdont forget to add --fast to server backend
Fit is FLux , in Comfy maybe is the workflow im using it goes faster on Forge, I'm too used to it by now
FI also do a bunch ois stuff with animation mostly
Fso, comfy all the way
PIs it possible to train against the Mochi preview models in Kohya? https://huggingface.co/Comfy-Org/mochi_preview_repackaged/tree/main/split_files/diffusion_models
Fi didnt see any work for it
Da part of this was merged into the Prodigy package and anyone can use it in other trainers
ZHi, could you tell me if I can train another character in DreamBooth using the same flux.dev model that was additionally trained on a different character? For example, I trained it on Arnold Schwarzenegger’s appearance, and now I want to create a dataset for Sylvester Stallone, or, for instance, train it in a specific style like Dark Fantasy.
