Great video man... Got it working. What would you suggest for people that have higher VRAM? Can this workflow be improved? Not the 72GB Vram scenario, but what about 24 or 48? Thanks!
Thanks for this walk through, I went through it all today and just got the comfyUI part working after doing the initial modeling work earlier today. Very good learning exercise for me
Earned the subscription, great work! I was wondering what would be the most effective system to have if I wanted to be able to generate an image that had N fine tuned objects or people? For example, lets say I trained it on me and my two brothers, and I wanted to generate an image with us three, is there an effective way to train a single model on 3 folders of different labels of images, or is there a better way by training three independent loras and combining them somehow? If you figured this out and made a full video it would be groundbreaking, unbelievably applicable. Loved the video, excited to see whats next!
You mention 26:00 that you've over-trained model. How would you change config (at the beginning of the video) to avoid that? Would adding more images reduce effects over-training? Thank for tutorial
Over training in this scenario isn't necessarily bad- I use this with varying CFG and lora weight scale to get consistent results. Other approaches would just be less steps of training
@AdamLucek So, by your opinion, do you recommend FLUX is better than SDXL or SD3 medium that its(Flux's) trained Lora model can generate more consistent portrait of myself? (with the same training images)
Thank you so much for your channel! I am learning so much! I have subscribed. I have a new Windows 11 machine with the following specs: System Manufacturer ASUS System Type x64-based PC Processor AMD Ryzen 7 7800X3D 8-Core Processor, 4201 Mhz, 8 Core(s), 16 Logical Processor(s) Installed Physical Memory (RAM) 32.0 GB Total Physical Memory 31.2 GB Available Physical Memory 1.46 GB I do not have any graphics card installed, using onboard graphics. My plan was to add a graphics card, but have not done so yet. The file nvidia-smi.exe is not on my PC anywhere, looked where various online resources said it would be and performed a search. I am showing hidden files if that makes a difference. How do I proceed from here? Thank again!
he's almost certainly using lambda labs. I fired it up today to try to go through this 'lab' as he does and the interface is 100% the same. a single A100 GPU node is only $1.29 USD per hour, pretty cheap
Great video man... Got it working. What would you suggest for people that have higher VRAM? Can this workflow be improved? Not the 72GB Vram scenario, but what about 24 or 48? Thanks!
Thanks for this walk through, I went through it all today and just got the comfyUI part working after doing the initial modeling work earlier today. Very good learning exercise for me
Great to hear that!
Earned the subscription, great work! I was wondering what would be the most effective system to have if I wanted to be able to generate an image that had N fine tuned objects or people? For example, lets say I trained it on me and my two brothers, and I wanted to generate an image with us three, is there an effective way to train a single model on 3 folders of different labels of images, or is there a better way by training three independent loras and combining them somehow? If you figured this out and made a full video it would be groundbreaking, unbelievably applicable. Loved the video, excited to see whats next!
You rock man, keep the great content coming!
Nice tutorial, thanks for sharing!
Thank you! Cheers!
Nice tutorial, thank you for sharing
You mention 26:00 that you've over-trained model. How would you change config (at the beginning of the video) to avoid that? Would adding more images reduce effects over-training? Thank for tutorial
Over training in this scenario isn't necessarily bad- I use this with varying CFG and lora weight scale to get consistent results. Other approaches would just be less steps of training
@AdamLucek So, by your opinion, do you recommend FLUX is better than SDXL or SD3 medium that its(Flux's) trained Lora model can generate more consistent portrait of myself? (with the same training images)
Flux for sure! But requires more compute
@@AdamLucek how if buy one more gpu card, my current one is 3090 ti
For your own machine, its usually recommended to have at least 24gb of VRAM for Flux training
Thank you so much for your channel! I am learning so much! I have subscribed.
I have a new Windows 11 machine with the following specs:
System Manufacturer ASUS
System Type x64-based PC
Processor AMD Ryzen 7 7800X3D 8-Core Processor, 4201 Mhz, 8 Core(s), 16 Logical Processor(s)
Installed Physical Memory (RAM) 32.0 GB
Total Physical Memory 31.2 GB
Available Physical Memory 1.46 GB
I do not have any graphics card installed, using onboard graphics. My plan was to add a graphics card, but have not done so yet. The file nvidia-smi.exe is not on my PC anywhere, looked where various online resources said it would be and performed a search. I am showing hidden files if that makes a difference. How do I proceed from here? Thank again!
Why do my images get horrendously stretched when using my newly trained Lora?
Can i use flux on my rtx3060 12gb vram?
Yes! The setup I have for generating images is on my RTX 3060 with 12gb vram, the training however requires atleast 24gb with this method
Is there a training you can do on mac? Or is it purely nvidia?
What cloud provider do you use to rent the A100 sir?
he's almost certainly using lambda labs. I fired it up today to try to go through this 'lab' as he does and the interface is 100% the same. a single A100 GPU node is only $1.29 USD per hour, pretty cheap
great
Flux dev is not free😅
Can u please do the "configuration file explanation & setup" part on ur pc just to show how it would work for beginners?