Atomix Pony XL
This is a mixed model based on Pony Diffusion V6 XL, which merged a variety of other models and Lora. Finally, I have achieved a style that I feel good about, I hope you will also like it.
Tips
This model is base on pony,so any pony Lora is working for this model.
This model has merged common Lora, which you can see from the Suggested Resources. So you do not need to import these Lora again.
Each image I upload contains a complete workflow, which you can download the image and drag to the ComfyUI interface to obtain.
Suggested Setting
Positive prompt:
score_9, score_8_up, score_7_up,Nagative prompt:
score_4, score_5, score_6, Sampler:(Turbo acceleration only work with these settings):
cfg: 1~2
steps: 6~10
sampler_name: dpmpp_sde_gpu
scheduler: sgm_uniformDescription
FAQ
Comments (15)
My Automatic does not have that sampler.. is it okay to use Euler A with normal CFG and Steps?
Your automatic should have this sampler and scheduler. It's DPM++ SDE and then select SGM Uniform from the Scheduler dropdown. If you haven't updated your automatic1111 you should do that; if you have then it sounds like your install is corrupted. If that's the case, then do a full reinstall.
To answer your question though, yes, you can do Euler A, Euler, DPM 3M SDE, and so on. Some will work better than others.
Euler A will work too.
Check this post with CFG and Samplers test:
https://civitai.com/posts/2687203?returnUrl=%2Fmodels%2F340158%2Fatomix-pony-xl%3FmodelVersionId%3D470632
Based on my tests so far: use Heun or DPM++ SDE (or DPM++ 2M Karras, there is not so much visual difference) if you need more sharp and detailed results, and Euler if you need to explore seeds (as it is the fastest so far, if not look at LCM). Then you can upscale or generate that seed again with DPM++ or Heun. Use Euler A or DPM2 A to explore even more variants, but you can only upscale from them. Generating the same seed with non 'a' (ancestral) will give you a different result.
Or, if you don't want to jump into all this stuff and just generate pictures occasionally, try to use Fooocus instead - simpler, close to Midjourney, but doesn't have some cool features like aDetailer, HiresFix, ControlNet, etc.
@LetTheBassDrop You are right I do have it!!! thank you!!
@Damacy Thanks a bunch! I got it working, it does clothing so pretty also!!
@EllaLove Also, advice to learn about prompting for SD. Try to use something like 'low angle' or 'high angle', 'fashion photography', 'glamour pose', 'fashion magazine', 'professional photo', etc. for more dynamic poses as SD is a bit stiff here. CivitAI usually shows generation data, but consider that final work may include Inpainting... sometimes a lot. Every model reacts differently to the same prompt. Some may generate more dynamic poses, others will be better with details. You also can generate with one model and then upscale(or refine) it with another one to get both, just set the Denoise to 0.6-0.75 to give that second model some space to work. A lot of possibilities... but also a lot of pain to learn and master this all to get the desired results.
Happy learning!
Once you dive into it... there is no way back🙈
Alex, mate! Congrats on a new release.
By the way, what kind of funky stuff did you do this time?
The main change is to change the style of the eyes,
And the overall image lines are softer and more anime style.
BEST MODEL OF PONY BASED EVER
Wow,That's really a high compliment.Thank you, mate.🤝
Look At Me, I'm The Checkpoint Now
This has become your stage, keep up the good work👏
Hi, I've been using this model in combination with openpose for creating my character palette (grid with different poses). For some reason, whenever the controller is enabled with my skeleton sheet, the output is always blank. I've been going through a lot of git links to find the root cause for this but haven't had any success. Anyone else who was able to get this done?
P.S this is what the error output says (not able to debug even after going through the code so any help is appreciated)
/webui_forge_cu121_torch21/webui/modules/sd_samplers_common.py:75: RuntimeWarning: invalid value encountered in cast | 11/50 [03:29<10:12, 15.70s/it]
x_sample = x_sample.astype(np.uint8)
hello, do you use high res at all?
if yes what model do you use?
Is a good model, my first model actually. so bad there is no more versions.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.