Wild_Child V3.0: I wanted to train this version to be "wilder" again, but with more sharpness and contrast. I hope you enjoy the results. If you are looking for consistency, I would recommend V4.0, but if you want a little chaos, stick with wild child.
V4.0: I used different methods to train this version to have a little more variety and sharpness. With trial and error, I got a nice clean look that I liked.
Wild_Child V2.0: This model is still wild, but maybe a little less so. I just wanted to keep the style of the original and add better prompt adherence.
V3.0: Combination of style and prompt recognition upgrades. Images come out very clean with a very unique style. Jack of all trades type of checkpoint. This is a continuation of V2.0, Wild_Child is still it's own thing.
Wild_Child: This model can be "wild" with how it responds to some prompts. The style is very unique and looks completely new. I like this one for when I want a certain look I can't get with any other model. Mild lighting and shadows. This model is more for playing around with and being surprised by results, such as abstract works.
Lightning models best results settings:
Sampler: dpmpp_2m_sde_gpu
Scheduler: sgm_uniform
CFG: 1.0-1.8
Steps: 10
V.4 and Wildchild settings I use comfyUI :
Steps: I use 28
cfg: I use 4.5
dpmpp_3m_sde_gpu
sgm_uniform or karras
Vae is baked in
Description
I wanted to train this version to be "wilder" again, but with more sharpness and contrast. I hope you enjoy the results.
FAQ
Comments (20)
Lucky for you, the most complex prompt had the desired effect.
Very nice model, almost impossible to produce something bad looking. Still, need to test LoRAs and stupid stuff though :)
I am glad to have you around to test it out.
As someone who loves your models,
I have a few questions.
Your models have delicate details and great color expression.
But because of that, the contrast sometimes gets burnt out.
This may be because I'm not good at controlling them,
But because of that, extensions like free U get in the way.
If there's a setting method that controls the colors well, can you tell me?
I don't really use Free U much, but for me the best way to control contrast other than in the prompt is to adjust CFG or trying different samplers and schedulers. I like to use dpmpp_3m_sde_gpu, sgm_uniform at CFG 4. I also toy around with different positive and negative embeddings and use different color and contrast prompting depending on the style I want.
Thank you very much for your answer.
The model's performance is so good that I have a desire for creative creations in many ways,
so I'm looking for various methods,
but I often ask because the color part is not well controlled.
I will try to do my best based on the ideal settings sampler, step count, and cfg scale posted on the homepage.
I'm sorry if my English is not good because I'm Korean.
If you use a1111 or similar UI (forge, reforge), there is an extension for that GitHub - Haoming02/sd-webui-vectorscope-cc: An Extension for Automatic1111 Webui that performs Offset Noise* natively.
I think there is something similar to that for comfy also. I hope this may help you.
@fertyt449107 Thank you. I will try.
Using your model a lot. Interestingly, I find to prefer the output of V3.0 vs. V4.0; for me, that model is far more creative in drawing detailed fantasy backgrounds and various magical items. V4.0, while probably more consistent, is a bit more boring for this style of prompts. (Haven't tried the "Wild Child" yet).
Amazing model
Thank you!
Lightning v3.0 is good.
Thank you, I am glad you like it.
Any chance of you making a flux model? Pretty please, puppy Eyes
check my workflow on openartai Shubz World
This is....Fantastic. Really. Thanks alot for, imho, the best Checkpoint there is for SDXL, still.
Wow, thank you for these kind words. It's pretty cool that my models are still being used and appreciated. Once things settle down a bit, I might dabble in creating more models.
pony please
This model is so fucking incredible! Well done! Just wowowowowowow!!!!!!!!!!!!!
By Far still the best Model for SFW , no doubts...
As I said before it's incredible but it would be really nice to include the generation data (minus the prompt) like Which Scheduler/Sampler/CFG/Steps for your images. ♥
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.