V8 is a mix of the past v7 v6 and v5.
V8 mix B has better knowledge of characters and need less weighting to get colored skin for example. Not that it is always correct but it has a better idea of who a character is.
Either way both can be interesting, curious which one you prefer.
Rule of thumb for both hyper versions, if you get artifacts either our cfg is too high or your steps too low. Hyper versions are Euler and Euler a only most other samplers don't really work.
Version 6 seems to behave with most samplers, not sure if I would recommend Euler a however.
Removed 5.1 it was too buggy so I removed it.
Not sure if Clip merging is such a good idea especially with Pony models.
Did remove the T5 clip merge as it looks like it does not work, or only has an effect with A1111. If you are using Comfy UI there is no difference between v4 with or without T5.
The real version 9 of my fast Pony merge.
I decided to publish this under a new name as it will where I try to get the look I want (more then in the other one where I tried to keep everything Pony intact).
I don't think I broke anything Pony like yet...
Will leave the name however as "fast" as it is based on version 7.5 and you can get decent results with 10-12 steps.
Don't use Euler a, especially with version 2 (at least for me it causes issues)
DPM ++ samplers are fine
I hope nobody minds the prompts I have been borrowing to se how mine compares.
Description
renamed to just v5 as it is not really T5 clip
merge between v4 t5xxl or v4 (which seem the same especially in Comfy) and fast pony merge v12 (which I could not decide if it was worth releasing), latest lora and a sprinkle of magic (a.k.a. another merge where I can't remember what I put in except my centerfold I think).
FAQ
Comments (9)
I am trying to use the method you outlined to merge the t5xxl encoder into my own custom Pony model, but am unable to do so with the FP8 encoder since I do not own a 4000 series GPU, and thus I cannot load any FP8 data types in ComfyUI.
In my testing with the FP16 version of the t5xxl encoder there is no difference in generation between a checkpoint with the t5xxl clip merged and the same checkpoint without; both images are identical.
So is it required for the merge add to use the FP8 version in order for the t5xxl encoder to have any effect?
@seto007 @TiwazM Yeah it's kinda strange. v40 and v40WithT5xxl gives me identical results in ComfyUI, but slightly different results in A1111, and there and v40WithT5xxl is 198 MB larger than v40. When I save copies of each using the checkpoint save node, they become the same size again.
@bot_i_celli my friend who was able to load the FP8 model ran the test and same result; identical images. We both tried using various different model merge nodes with different weights, and none of them seemed to have any impact whatsoever.
Starting to wonder if this is just a case of confirmation bias and the t5 model isn't getting loaded with SDXL at all, or if there's something we're missing.
Interesting, I thought it was strange that it sometimes worked and others not but I only tested them in A1111. I will have a look in Comfy.
I used the method outlined in the model desc (ClipMergeAdd) to load a different Pony checkpoint w/ t5xxl and saw no difference in output...
It definitely seems like this method of "adding" t5xxl to SDXL/Pony checkpoints might not be working properly, lol
what is a good link to the clip sources? thanks and great work!
Great model, thx so much!
Having a lot of fun with this model, you can really let your imagination run wild and see amazing results. Having problems with 5.1 version no matter what I do I get overbearing artifacts.
After some more testing I noticed some issues too will see if I can fix it, maybe I should remove it for now.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.