Experimenting with different settings to get better realism with Flux, what are your secret tricks?
https://redd.it/1lidhjy
@rStableDiffusion
https://redd.it/1lidhjy
@rStableDiffusion
Reddit
From the StableDiffusion community on Reddit: Experimenting with different settings to get better realism with Flux, what are your…
Explore this post and more from the StableDiffusion community
🔥 90% OFF - Perplexity AI PRO 1-Year Plan - Limited Time SUPER PROMO!
https://redd.it/1lihgxv
@rStableDiffusion
https://redd.it/1lihgxv
@rStableDiffusion
1 big LoRA or 4 small LoRAs? (Same item, different styles/contexts)
I have 4 datasets of the same relatively simple object (corporate logo), but in 4 different contexts that change its shape:
one is printed on a t-shirt (various people wearing these shirts in different conditions)
one is on a coffee mug (various mugs in various conditions)
one is as a flat advertising banner (but there's not much variety in visuals)
one is of the object stylised in a variety of marketing cartoons/material (in every style imaginable from 2D to 3D)
Each dataset is over 100 images, and has good quality captions, with a consistent captioning style. Using OneTrainer on SDXL at the moment.
My attempt at a combined LoRA was to have captions like "c0rp0logo on tshirt" or "c0rp0logo on mug". It struggles to get good results. I guess it is because there is some context bleeding between these different contexts. But I'm a noob so I could easily just be missing something obvious - I've only trained a dozen or so LoRAs of individual items before.
But the combined LoRA is preferable because we might want to put this logo on other types of object!
Is there a way to get it to work as a single combined LoRA? What do you think I might be getting wrong?
https://redd.it/1lihi1p
@rStableDiffusion
I have 4 datasets of the same relatively simple object (corporate logo), but in 4 different contexts that change its shape:
one is printed on a t-shirt (various people wearing these shirts in different conditions)
one is on a coffee mug (various mugs in various conditions)
one is as a flat advertising banner (but there's not much variety in visuals)
one is of the object stylised in a variety of marketing cartoons/material (in every style imaginable from 2D to 3D)
Each dataset is over 100 images, and has good quality captions, with a consistent captioning style. Using OneTrainer on SDXL at the moment.
My attempt at a combined LoRA was to have captions like "c0rp0logo on tshirt" or "c0rp0logo on mug". It struggles to get good results. I guess it is because there is some context bleeding between these different contexts. But I'm a noob so I could easily just be missing something obvious - I've only trained a dozen or so LoRAs of individual items before.
But the combined LoRA is preferable because we might want to put this logo on other types of object!
Is there a way to get it to work as a single combined LoRA? What do you think I might be getting wrong?
https://redd.it/1lihi1p
@rStableDiffusion
Reddit
From the StableDiffusion community on Reddit
Explore this post and more from the StableDiffusion community