rule of thumb is your Lora must be rank 128 and trained on high quality realistic images
rule of thumb is your Lora must be rank 128 and trained on high quality realistic images
Afinetuned checkpoint even better but expensive
HI'm using civit loras
Ayeah not gonna work
HThey all have high ratings, what am I missing?
Awell yeah but there usually rank 16 -64, at best
AI suggest following doctors tutorials
HSo I understand what you're saying, but if I see plenty of people with really nice results, I am wondering.. What in my settings am I missing? Everybody else is getting nice results but somehow some of my gens are not. That's my question
Acan you provide an example
HAthis seems just a high resolution image to me not better realism
Aalso try to play with lightning and textures in your prompt
Areal realism always comes with a bit of noise too (cause reality ain't perfect)
HWhat about this workflow ? I dont see schedulers
H
Fi recommend to use swarmui
HUnfortunately itll be dated soon, comfy is the best way
Fswarmui using comfyui
RDo you have examples to support this claim? Or in what sense you consider Onetrainer to be superior to Kohya?
HMore intuitive
HAnyone ever make a celeb into a baby style lora? I can't find it anymore, but someone made like a baby Trump for example, thats funny I would like to try this too
K
Twhich one you suggest?
Tis there any truly better than BFL dev?
M@Dr. Furkan Gözükara How much vram do i need to use that text + image to video models ?
Nis it possible to change the generation output folder to a different path?
e.g. now c/Users/User/AppData/Roaming/StabilityMatrix/Packages/Stable_Diffusion_WebUI_Forge2/output/
and need c/SD_output
e.g. now c/Users/User/AppData/Roaming/StabilityMatrix/Packages/Stable_Diffusion_WebUI_Forge2/output/
and need c/SD_output
Mwhat model si this?
M@Dr. Furkan Gözükara and isnt there any model available that can generate videos upto 1 minute or maybe 30 seconds ?
Fi dont know you need to test. you can look civitai
Fwhich model? give me name
Fi dont use stability matrix and not recommending either 
it may cause errors
it may cause errorsFnot yet but i am following a repo
Fwhich can double duration
Falso i am removing duplicate model issue from our wan app right now
Falso adding squish effect lora
Tsome of them are just bad lora merges. results are more often than not worse than stock flux dev
Tnot sure if it can be set in config
but if all else fails:
https://schinagl.priv.at/nt/hardlinkshellext/hardlinkshellext.html
check this tool .
but if all else fails:
https://schinagl.priv.at/nt/hardlinkshellext/hardlinkshellext.html
check this tool .
Tbtw stability matrix is really good. i like it alot
FFsquish lora working
Fi am adding to our downloader
F
NMatrix is convenient in its own way, easy to follow multiple interfaces, easy to roll back, easy to add new models
I've been using it for a year or more
I've been using it for a year or more
XWow this looks so much better than the pika.labs squish generated ones
Fi made this :d
Fi will make a quick video
Fworking on removing duplicate models
Fwith backwards compatability
Ftesting new downloader
F
Fit will save over 50 GB when all models downloaded