Search Results
7/26/2025, 1:04:49 AM
>>511367576
I am white though
I am white though
7/25/2025, 2:52:44 PM
Weren't you guys banned from this website by your nanny state?
7/21/2025, 9:30:51 AM
>>212986980
non ho capito
non ho capito
7/2/2025, 5:55:02 AM
Is it just me or is most band merch really ugly? And why do they always put the cool designs on the back of the shirt?
6/24/2025, 5:30:09 PM
I've been following along with recent model developments in both NAI and the finetunes, and I just want to ask:
Has any research or work gone into making characters and backgrounds more consistent across separate img gens?
Are the only solutions training a lora, reference only controlnet, and inpainting like a manual painter to fix inconsistencies?
Something as simple as a comic with the same character across multiple panels and pages easily exposes even the "best" models as "slop".
The image sets I've seen that are consistent are all the most normie popular characters and lack distinct details so are consistent because there are no finer details to be inconsistent on.
Above mentioned current solutions that I know of all kinda fail when it comes to detailed characters too.
With the advent of image generation and multi-frame consistency, do we still not have something for image gen that takes in one reference image and consistently generates the same character in different poses or same background with different characters?
In theory, should be similar to img2vid.
Some examples of this inconsistency are like a guy in a suit, sure I can prompt for a blue tie every time, but upon closer inspection, one image will have a tie clip and the next won't. One image will have 2 buttons next will have 3 buttons. Small stuff like this that immediately shows it's generated and not drawn by a human who would take into consideration these simple details.
Is there a newer better base model or plugin or workflow or whatever that solves this?
Has any research or work gone into making characters and backgrounds more consistent across separate img gens?
Are the only solutions training a lora, reference only controlnet, and inpainting like a manual painter to fix inconsistencies?
Something as simple as a comic with the same character across multiple panels and pages easily exposes even the "best" models as "slop".
The image sets I've seen that are consistent are all the most normie popular characters and lack distinct details so are consistent because there are no finer details to be inconsistent on.
Above mentioned current solutions that I know of all kinda fail when it comes to detailed characters too.
With the advent of image generation and multi-frame consistency, do we still not have something for image gen that takes in one reference image and consistently generates the same character in different poses or same background with different characters?
In theory, should be similar to img2vid.
Some examples of this inconsistency are like a guy in a suit, sure I can prompt for a blue tie every time, but upon closer inspection, one image will have a tie clip and the next won't. One image will have 2 buttons next will have 3 buttons. Small stuff like this that immediately shows it's generated and not drawn by a human who would take into consideration these simple details.
Is there a newer better base model or plugin or workflow or whatever that solves this?
6/13/2025, 2:43:57 AM
>>507151652
What did Brazil ever do to you?
What did Brazil ever do to you?
Page 1