- Sep 15, 2021
- 1,047
- 4,002
Quick FYI.
In case you were unaware,
Some very interesting ones for us fans of the DCAU.
The built-in Poison Ivy looks pretty good:
In case you were unaware,
You must be registered to see the links
is a list of all characters pre-baked into the Pony Checkpoint (and thus all of its derivatives) - no LoRA required.Some very interesting ones for us fans of the DCAU.
The built-in Poison Ivy looks pretty good:
- Dec 23, 2018
- 598
- 2,511
Yeah inbuilt pony characters are quite good. Here's a few I did with fennfoto realistic pony checkpoint just with pre-baked stuff
Also I'm sure you all use wildcards, so here's a couple I got from a Pony discord server showing what you can produce
The anime list is long also.
You don't have permission to view the spoiler content.
Log in or register now.
You don't have permission to view the spoiler content.
Log in or register now.
You don't have permission to view the spoiler content.
Log in or register now.
You don't have permission to view the spoiler content.
Log in or register now.
- Sep 15, 2021
- 1,047
- 4,002
This young lady is starting to annoy me now!
She's frustratingly close to perfect. Unfortunately, I get the teeth right and the hands go wonky, I get the hands right(ish) and the nipples get black spots. I've probably got too many steps, too many prompts, gawd knows what else.
Any advice (other than telling me to use CUI!
)?
N.B. I'd like to keep the same Checkpoint if possible to retain her look. I also want a Pony-derived one so I can more easily have her do naughty shit later!
Here are three consecutive seeds, I hope you can see why I want to get her 'just right'...
(JPEGs posted rather than PNGs for size)
She's frustratingly close to perfect. Unfortunately, I get the teeth right and the hands go wonky, I get the hands right(ish) and the nipples get black spots. I've probably got too many steps, too many prompts, gawd knows what else.
Any advice (other than telling me to use CUI!
)?N.B. I'd like to keep the same Checkpoint if possible to retain her look. I also want a Pony-derived one so I can more easily have her do naughty shit later!
Here are three consecutive seeds, I hope you can see why I want to get her 'just right'...
(JPEGs posted rather than PNGs for size)
You must be registered to see the links
Got tired of seeing Maui showing up in my AI images also being the most popular HDR for Daz
You must be registered to see the links
Last edited:
Hey, im not sure if asking questions over here is the right way to do so.
I was wondering for a while now, since im planning on making a short/light hearted game. Always wanted to create one but never had any experience in art.
Now here comes AI, i've been using SDXL for quite sometime now. Been experimenting with ComfyUI and the results were 10/10.
The problem is, i dont know how to get consistent characters (at least consistent faces ~80% similarity). Is Lora really the way to go ? or should i just use Pony checkpoints and use characters that are available there. Would be nice if i can create my own character and then train them because i've fallen in love with Dalcefo's models (i've attached one of his old models based on sd1.5)
Also, i've been using cloud GPU all this while (24gb vram. L4)
Basically, is there any way to get consistent characters with at least 80% similarity.
I was wondering for a while now, since im planning on making a short/light hearted game. Always wanted to create one but never had any experience in art.
Now here comes AI, i've been using SDXL for quite sometime now. Been experimenting with ComfyUI and the results were 10/10.
The problem is, i dont know how to get consistent characters (at least consistent faces ~80% similarity). Is Lora really the way to go ? or should i just use Pony checkpoints and use characters that are available there. Would be nice if i can create my own character and then train them because i've fallen in love with Dalcefo's models (i've attached one of his old models based on sd1.5)
Also, i've been using cloud GPU all this while (24gb vram. L4)
Basically, is there any way to get consistent characters with at least 80% similarity.
- Dec 23, 2018
- 752
- 575
Lora definitely works, but there're lots of nuances. Your question sounds a bit like "Basically, is there any way to make a porn game with programming with at least 80% success?"
Well, i guess the answer is Lora huh. Looks like i have to really look into loras. Dont get me wrong, this game is solely a hobby/side project because i think its a good way to find like minded people and maybe make a new community from it
- Dec 23, 2018
- 752
- 575
One of the tricks, for example, is using Adetailer (whatever there're alternatives for comfyui) and impainting/detailing the face with your lora, while the other image is generated without this face lora.
Another trick is making several loras (separately for the same character and for his/her clothes, etc) for several sdxl and sd1.5 models, and using even more complicated pipe in comfy, etc.
It's just you can start looking into lora, and abandon your wish to make any game, because you'll be too busy with generating images. Happens too often...
I actually like your idea of making a game with the oil paintings of elven maidens, meanwhile.
Last edited:
Lora if your trying to make a "new" character is an option.Well, i guess the answer is Lora huh. Looks like i have to really look into loras. Dont get me wrong, this game is solely a hobby/side project because i think its a good way to find like minded people and maybe make a new community from it
Another is to blend celebrity's but that would be less consistent. [celeb A: celeb B: 0.5]
Art style is another thing, if you want consistency their.
Just keep in mind even a short VN would have few hundred images.
With Pony I chose a very well trained character (Korra) and I have been working on and off with that for awhile.
Its little things like a tan line being slightly off or a tongue merging into a penis.
Fixing things like that can take a few minutes per image over 100's of images adds up.
- Sep 15, 2021
- 1,047
- 4,002
Another suggestion, assuming you don't want an existing character, is to just put a name in the prompts. Not a real person, just a name that would be plausible for the character.
I can't post any examples as I'm away from my PC atm, but I wanted to create a consistent face for an AI supermodel.
So I just used 'Helena Rutakis' in the prompt, among everything else, as if it were a celebrity name or an Embedding. The results were pretty consistent - but bear in mind this is across just a single checkpoint. Change the checkpoint and you'll get a new face (albeit just as self-consistent).
Adapt-Lion + Full FP8
For anyone that does training I have found that I can run a full FP8 training using adapt-Lion on XL and PONY
Consider I only have an 8GB card so this is pretty impressive.
Its running 3-5x faster then Adam8bit
I think it would be possible for a 24GB Card user to Dreambooth or Full Finetune if they ever do the scripting for FP8
I am not using Deepspeed or Dynamo the only thing I have custom is a build of 12.5 Cuda with CudNN + CuBlas
The CuBlass might just be taking up 1GB of extra VRAM honestly
For anyone that does training I have found that I can run a full FP8 training using adapt-Lion on XL and PONY
Consider I only have an 8GB card so this is pretty impressive.
Its running 3-5x faster then Adam8bit
I think it would be possible for a 24GB Card user to Dreambooth or Full Finetune if they ever do the scripting for FP8
I am not using Deepspeed or Dynamo the only thing I have custom is a build of 12.5 Cuda with CudNN + CuBlas
The CuBlass might just be taking up 1GB of extra VRAM honestly
- Dec 23, 2018
- 598
- 2,511
You must be registered to see the links
The new Flux model is out and is very impressive. Unfortunately needs 12+ gig of vram to work (although some claims that it runs on 8gig). It's alreaady working in comfyui, not sure about auto1111
Examples -
Last edited:
- Mar 3, 2019
- 237
- 775
Anyone wanting to play around with this model but lacking in hardware requirements, there is a free website using its API that you can create images. Only needs a google account for login.
You must be registered to see the links
- Mar 3, 2019
- 237
- 775
From the way it's written, I would think it's the other way around, no? If you select "no sysmem fallback", it will disable the ability to use the system memory as fallback when there isn't enough VRAM. If you select the last option, "prefer sysmem fallback", it should allow that to happen. At least that's my understanding.
- Dec 21, 2020
- 1,571
- 3,768
ComfyUI noob FLUX workflows:
To get one started with the least demanding model (schnell), one needs a few files: the "vae", the model, the workflow, and the encoders.
The "vae" (ae.sft), the model (flux1-schnell.sft) are here:
The encoders:
The very first link here explains where each file goes.
Looks like a cool 40GB in new downloads when all said and done.
The workflow for schnell is last one in this table:
You must be registered to see the links
To get one started with the least demanding model (schnell), one needs a few files: the "vae", the model, the workflow, and the encoders.
The "vae" (ae.sft), the model (flux1-schnell.sft) are here:
You must be registered to see the links
The encoders:
You must be registered to see the links
The very first link here explains where each file goes.
Looks like a cool 40GB in new downloads when all said and done.
The workflow for schnell is last one in this table:
| Dev | FP8 | Schnell |
|
|
|
Last edited:
- Dec 21, 2020
- 1,571
- 3,768
Yea, I dunno, I keep getting error "!!! Exception during processing!!! Error while deserializing header: InvalidHeaderDeserialization" when trying the FLUX workflow.
Can't immediately troubleshoot what's up
Can't immediately troubleshoot what's up
You don't have permission to view the spoiler content.
Log in or register now.
Someone merged the
17 and 21 GB so if you have a 3090 or 4090 your golden.
For me its around 100 seconds per IT so I am out on using it.
Make sure your not using any crazy launch arguments like --force-fp32
If your AMD and using --directml, I don't think Olive has been updated to convert the model yet
I'm not sure if the defualt XFormers or --use-pytorch-cross-attention would be faster.
I also built CUPY but im not sure it matters it adds in CuBlass, part of my mission to get Flash Attention, Cutlass and Deepspeed working together.
It would be my only shot at using Flux at a decent speed
You must be registered to see the links
into FP8, you need to use the newest version of COMFY17 and 21 GB so if you have a 3090 or 4090 your golden.
For me its around 100 seconds per IT so I am out on using it.
Make sure your not using any crazy launch arguments like --force-fp32
If your AMD and using --directml, I don't think Olive has been updated to convert the model yet
I'm not sure if the defualt XFormers or --use-pytorch-cross-attention would be faster.
Did you pull the latest version of COMFY, I had a float error until I updated.Yea, I dunno, I keep getting error "!!! Exception during processing!!! Error while deserializing header: InvalidHeaderDeserialization" when trying the FLUX workflow.
Can't immediately troubleshoot what's up
I also built CUPY but im not sure it matters it adds in CuBlass, part of my mission to get Flash Attention, Cutlass and Deepspeed working together.
It would be my only shot at using Flux at a decent speed
Last edited: