Even_Adder@lemmy.dbzer0.com to Stable Diffusion@lemmy.dbzer0.comEnglish · 13 days agoStable Diffusion 3 Medium Fine-tuning Tutorial — Stability AIstability.aiexternal-linkmessage-square8fedilinkarrow-up111arrow-down11
arrow-up110arrow-down1external-linkStable Diffusion 3 Medium Fine-tuning Tutorial — Stability AIstability.aiEven_Adder@lemmy.dbzer0.com to Stable Diffusion@lemmy.dbzer0.comEnglish · 13 days agomessage-square8fedilink
minus-squareistanbullu@lemmy.mllinkfedilinkEnglisharrow-up2·12 days agoIs a reason to use SD3 when Flux is much much better, with a much better license?
minus-squareEven_Adder@lemmy.dbzer0.comOPlinkfedilinkEnglisharrow-up2arrow-down1·12 days agoI don’t think so. They’re going to have to do a lot better than a tutorial to win people back.
minus-squareistanbullu@lemmy.mllinkfedilinkEnglisharrow-up1·12 days agokohya now supports flux fine tuning. I have seen nice examples in civitai.
minus-squareEven_Adder@lemmy.dbzer0.comOPlinkfedilinkEnglisharrow-up1arrow-down1·12 days agoThose might just be LoRA merged models, not full fine-tuning. From what I heard, fine-tuning doesn’t work because the models are distilled. You’d have to find a way to undistill them to train them.
minus-squareclb92@feddit.dklinkfedilinkEnglisharrow-up1·edit-212 days agoPeople have been training great Flux LoRAs for a while now, haven’t they? Is a LoRA not a finetune, or have I misunderstood something?
minus-squareEven_Adder@lemmy.dbzer0.comOPlinkfedilinkEnglisharrow-up1arrow-down1·12 days agoLast I heard, LoRAs cause catastrophic forgetting in the model, and full fine-tuning doesn’t really work.
minus-squareclb92@feddit.dklinkfedilinkEnglisharrow-up2·12 days agoOh well, in practice I’ll just continue to enjoy this (possibly forgetful and not-fully-finetunable) model then, that still gives me amazing results 😊
minus-squareerenkoylu@lemmy.mllinkfedilinkEnglisharrow-up1arrow-down1·edit-210 days agoquite the opposite. Lora’s are very effective against catastrophic forgetting, and full finetuning is very dangerous (but also much more powerful).
Is a reason to use SD3 when Flux is much much better, with a much better license?
I don’t think so. They’re going to have to do a lot better than a tutorial to win people back.
kohya now supports flux fine tuning. I have seen nice examples in civitai.
Those might just be LoRA merged models, not full fine-tuning. From what I heard, fine-tuning doesn’t work because the models are distilled. You’d have to find a way to undistill them to train them.
People have been training great Flux LoRAs for a while now, haven’t they? Is a LoRA not a finetune, or have I misunderstood something?
Last I heard, LoRAs cause catastrophic forgetting in the model, and full fine-tuning doesn’t really work.
Oh well, in practice I’ll just continue to enjoy this (possibly forgetful and not-fully-finetunable) model then, that still gives me amazing results 😊
quite the opposite. Lora’s are very effective against catastrophic forgetting, and full finetuning is very dangerous (but also much more powerful).