My gpu have 8 gb vram
My gpu have 8 gb vram
M
GitHub
[SIGGRAPH Asia 2023] Rerender A Video: Zero-Shot Text-Guided Video-to-Video Translation - GitHub - williamyang1991/Rerender_A_Video: [SIGGRAPH Asia 2023] Rerender A Video: Zero-Shot Text-Guided Vid...
Fhow many unique apparel item you have?
F8 gb is not good to do training
Fyour best option is kaggle SDXL dreambooth for free
Fye solving their repo isntall hard
Fuse my auto installer
Fit works
NI have 5000 for now. I wanted to experiment on this first.
NHave an L40 GPU with 48gb vram
Fhow are you planning to find 5000 uniqu captions haha :d
NPlanning to add new tokens to the clip tokenizer, found an option to do that.
Finteresting where did you find it?
Fit literally tokenize everything
Npipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")
pipe.tokenizer.add_tokens(list_of_tokens))
pipe.text_encoder.resize_token_embeddings(len(pipe.tokenizer))
pipe.tokenizer.add_tokens(list_of_tokens))
pipe.text_encoder.resize_token_embeddings(len(pipe.tokenizer))
Nlike this
Fwow you gonna train with diffusers good luck 
Nsince everydream2 uses diffusers i can simply put these lines in the pipeline
Fit is a very hard job to get it right
Ni am using everydream2
Fi heard it makes more sense
Fi can also give you private consultation for different approach. but it wont be able to bring you certain product
Fit would be more general one
Flike type of clothing etc
Nthat would be helpful
Nbut is it worth trying this approach?
Fi dont think so
Ffirst of all assigning 5000 unique tokens
Fwould take huge time
Fand you dont know if will work or not
Falso 80k images you really should use multi gpu
Fyou can run on runpod really cheap
For you can wait days :d
Nyeah planning to use multi gpus. I have unique ids for each product so planning to put some random alphabets that is not in the clip vocabulary. I have tried this with 500 products and I got decent results. my concern will the model remember all the tokens when I scale it to 5000
Fwith sufficient number of epochs i think should
PStill not seeing anywhere near a 70% increase.
25% increase, yes.
70%, no.
How are you equating this percentage ratio?
25% increase, yes.
70%, no.
How are you equating this percentage ratio?
Fwhat is your gpu
Fi am installing tensoRT right now on my new model
P4090
Vi have a ComfyUI workflow question. anybody have a good method of inpainting large images so that the sampler only works on the masked area?
example, i arrive at a large-ish image (3000x3000) by upscaling. I need to fix parts. if i use the Mask Editor and push it through the KSampler (VAEencode > Latent Noise Mask > KSampler) it seems it still works on the entire image rather than just the masked area. In A1111, I recall it just working on the masked area.
How would this be done in ComfyUI?
example, i arrive at a large-ish image (3000x3000) by upscaling. I need to fix parts. if i use the Mask Editor and push it through the KSampler (VAEencode > Latent Noise Mask > KSampler) it seems it still works on the entire image rather than just the masked area. In A1111, I recall it just working on the masked area.
How would this be done in ComfyUI?
Fhere my difference
F
F55%
Fbut there are so many other apps running
Ftherefore not max
Fperfect face quality and resemablence
BGot a link to the latest SDXL kohyah config training preset?
BLORA training that is
Fi prefer dreambooth sdxl
Fthen extract lora
