Package designed for versatile applications, leveraging different LoRa or TI technologies to ensure consistent performance.
Basically, the intention of this pack is to share what I use.
POST GRID EXAMPLES: https://civarchive.com/posts/324343
PACKAGE
Checkpoint SD 1.5: https://civarchive.com/models/87371
Embedding NEGATIVE: https://civarchive.com/models/87375
Lora PHOTO: https://civarchive.com/models/87376
Lora DETAIL: https://civarchive.com/models/87378
Lora Dark Theme: https://civarchive.com/models/87575
VAE RAW: https://civarchive.com/models/88390
My settings are typically as follows:
Sampler: DPM++ 2M Karras
Samples: 30
Width-Height Ratio: 512x768
CFG Scale: 5 - 6
Lora DETAIL: Recommended: 0.3 to 0.8.
Lora PHOTO: Recommended: 0.5 to 1.0.
Lora DARK THEME: Recommended: 0.35 to 0.7
Nice Start: <lora:difConsistency_photo:0.5>, <lora:difConsistency_detail:0.35>, <lora:difConsistency_darktheme:0.35> Neg: difConsistency_negative_v2
NOTE: I'm manually adding the prompts to all the images... patience is key.
Description
FAQ
Comments (16)
Thanks , will you be uploading the lora files also ?
Looking great.
Can you please write a short guide how to use them? Are they good to use with controlNet in txt2img?
I'm adding the prompts to the images. I use it with ControlNet, and it works well for me.
@rMada THank you so much for the feedback. I will definitely try this and share the results on the Gallery.
@NeoAnthropocene Which ControlNet model you trying? Is it openpose ?
@tulantuna1 I don't try it with anything yet.
I guess people are using it with controlNet <Reference>
Hey, thank you for this nice pack.
Quick questions as im a little confused, are they all baked into the checkpoint or you need to download each separately ?
And what is the rule with multiple loras? Can you add as many as you want ?
thanks
UPDATE: you have to download separately :)
You can use everything independently and combine the different LoRa and TI components as needed. The term "pack" can be confusing.
I just spent an hour looking through your difConsistency stuff. You are really doing some important work here in terms of making tools to control visual style. And you have an intuitive feel for how to direct the genie in the black box that makes this all work. I like a model that can make a 3-breasted woman probably as much as the next guy, but what you are doing is on a whole different level. Please keep at it!
The word "BREAK" at the beginning of the negative prompt sounds to me like the reserved keywords in programming languages. What effect does it have on stable diffusion, and do I need to add an extension to automatic1111 in order to use those keywords?
I seem to remember seeing an extension that could split a prompt into distinct parts to get a bit more control over image generation, and it used the "AND" keywords, and a few others.
Thanks in advance and best regards.
The word "BREAK" in the negative sense doesn't make sense. I use styles and apply a break everywhere, but in this case, it has no effect
Hi, may I ask why two heads are always generated? The details are fine, it's just that the images always generate two or more heads~~~I can't fix it with the special negative words
In high resolutions, you need to use HiRes. fix. In low resolutions, I don't experience that issue
Broken anatomy with "amputated" limbs, this checkpoint is like back to the beginning of SD1.5.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.