you can try training on a a100 to see if results are good first
you can try training on a a100 to see if results are good first
Fnot needed
Fit works perfect on A6000
Ffused back pass reduced VRAM usage by far
Falso there are block swapping
Fso it will work on 24gb too
Fbut fp8 didnt make any diff
Mbut very slow though right?
Ffor some reason
Fsame speed as lora
Mwith block swapping?
Fno without
Fwith block it will be slower i expect
F
Mbut it wont fit on 24gb without block swapping i think
Fwith fp8 it should
Fbut it didnt work 
Myeah the fp8 models arent making any difference not sure why
Fjust reported it as a bug to the kohya
Fit should reduce vram to fit into 24 gb
Fcpu offloading almost there
Mwhat kohya said
If you specify fp8_base for LoRA training, the flux will be cast to fp8 from bf16, so the VRAM usage will be the same even with full (bf16) base model.**
If you specify fp8_base for LoRA training, the flux will be cast to fp8 from bf16, so the VRAM usage will be the same even with full (bf16) base model.**
Dkohya now has fused back pass?
Dwhat did you have for Lora? I'm getting 5.7 s/it using OT, bs1 adafactor fused backpass
FResults from trying to train a character + a puppet mascot character
Fhe said fp8 doesnt work
Ffor fine tuning yes
Fworks amazing
Flooks like my checkpoints totally cooked
F
Fi guess i need to lower LR a lot
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=40)
Dgetting shredded
FHas anyone done tests regarding interactions between multiple LoRAs, or is there documentation somewhere how it works in detail (assuming it doesn't get so technical it goes over my head)
For instance, does LoRA load order matter? Is it better to train a model with two characters (like the one I did above with a character + mascot) compared to training two different LoRAs and using them both at the same time?
For instance, does LoRA load order matter? Is it better to train a model with two characters (like the one I did above with a character + mascot) compared to training two different LoRAs and using them both at the same time?
Fit doenst work well
Fyou have to train both in same image usually
AMetal slug 
AIs there any concensus on what is causing the grid lines/scan lines in some Flux LoRA's? I've read anything from: VAE issue, CUDA version issue, a bug in kohya, anti-AI watermark in training dataset. Has anyone successfully narrowed down what was causing it?
DI don't see anything here, but I've seen grid lines when I've tried to use a Lora at very high weights, like 2
Fit happens to me when i generate high res images
V@Dr. Furkan Gözükara good afternoon, as to your liveportrait, could I put a video and image in a ratio 3:4, and get generated video also in 3:4?
DThere was so much fuss about the GGUF models/ people running out of VRAM/ Flux being too large, that I spent the entire day today with my home PC (12GB 4070 Super) & ComfyUI, trying various models. To my surprise, I found:
1) ComfyUI could load and generate with the original 23GB Flux Dev Model (reasonable speed) on my PC & FP8 T5XXL. I do not understand how people are running out of VRAM? It seems both swarm UI and comfy ui have some resource management going on inside which lets you use the dev model. Am I missing something? Since it was working, I loaded Florence2 and SAM2 as well, ended up generating + inpainting with relative ease.
2) As for the GGUF models, I actually found that they took much, much longer than the 23GB model to generate images!! I tried Q6_K and Q8_O and was able to do the above operations with them as well. The quality was good, but obviously not better than the full size dev model.
3) I found that Swarm UI is somehow doing a much better job at generating with the LoRAs+Img2Img. On the other hand, it seems SwarmUI can't work with Controlnets+LoRAs at all. The results are laughable tbh. The inpainting on ComfyUI is the best!
1) ComfyUI could load and generate with the original 23GB Flux Dev Model (reasonable speed) on my PC & FP8 T5XXL. I do not understand how people are running out of VRAM? It seems both swarm UI and comfy ui have some resource management going on inside which lets you use the dev model. Am I missing something? Since it was working, I loaded Florence2 and SAM2 as well, ended up generating + inpainting with relative ease.
2) As for the GGUF models, I actually found that they took much, much longer than the 23GB model to generate images!! I tried Q6_K and Q8_O and was able to do the above operations with them as well. The quality was good, but obviously not better than the full size dev model.
3) I found that Swarm UI is somehow doing a much better job at generating with the LoRAs+Img2Img. On the other hand, it seems SwarmUI can't work with Controlnets+LoRAs at all. The results are laughable tbh. The inpainting on ComfyUI is the best!
TSo I am pulling my hair out (and I am already bald). I really appreciate all the work put in to the tutorials and I have followed each step. I am now getting an error that shows "13:11:50-486153 ERROR LoRA type must be set to 'Flux1' or 'Flux1 OFT' if Flux1 checkbox is checked." Flux is auto checked as the model is in there and loaded. Please let me know what I can do to fix this.
Tnvm! I think I got it.
Fyes it should work
F1 , 2 , 3 are all correct :d
Fdid you solve, are you using our config auto installer?
Fwaiting new flux trainings to be completed
F
Dmore? lol
![Dazzastrous [4090]](https://cdn.discordapp.com/avatars/871389071686135838/c4a5f4d81d9eeb18d856295d692e2c07.webp?size=16)
Fyep
Flots of more trainings i need to do