thank u very much 4 ur kind comment ☺

I am happy if u like EVO :)

thank u very much 4 ur kind words ♥

means alot ☺ I am happy u like Evo ^.^

thank u very much 4 ur kind words ♥ I am very glad u like Evo ^-^

Tinkering in a colorful playground: lots of flavors, lots of variety☺ I'm glad you like Lvl 2

ur using an extra VAE maybe?

@Ellie2024 

So regardless of whether it’s Forge, Comfy, or Onsite, the image should normally have a very similar output.🤔
My guess is that an additional VAE might be causing the difference ~ please avoid adding extra ones, since the model already includes one. 🙂

The showcase is set up so the images look almost identical to hires on site. That’s why I chose an upscale factor of 1.526 ^^
But even without hires, the results should look pretty similar.

If it’s not the VAE, then my next thought would be the denoising setting. The higher that value is, the more the output will deviate from the original seed.

@Ellie2024 Has it worked now? :)

I use 0,2 mostly to keep the look :) sometimes 0,4. Depending on the upscale factor, the look can otherwise appear stretched

@Ellie2024 

May I ask if you have already tried using my prompt with my settings without an extra VAE or Lora? I tested it again yesterday and got exactly the same image. On the site it looks slightly different but keeping the look, but that’s only because I’m using Forge, which handles the seeds a bit differently than on-site. However, that’s independent of the model itself.

@Ellie2024 hm :I I’ve tested this multiple times myself, and both on site and locally the results look the same.
That’s why I assume the differences come from individual settings or extra components (like VAE, LoRA, or the denoise workflow) rather than from the model itself.
I only use hires with 1.526 and 0.2 denoise locally to stay close to the on-site settings, so the results are basically identical in both cases ~ just like in my showcase. 🙂
Even without hires, the image still looks stylistically the same as in my showcase.
To show exactly what’s possible, I keep my prompts visible so anyone can easily follow and understand the process.

At this point I really can’t add anything further ~ I don’t know your hardware or the base settings of the program you’re using, so I can’t tell what else might cause the difference.
Since I’ve tested everything on my side and the results are consistent, I’d kindly ask you to check your own setup in detail. To be honest, I'm not sure what else could be the reason. Thanks for understanding.

Sadly I can’t share workflows since I’m on Forge and not Comfy locally :( but to make up for it I did some extra testing ~ even without hires or adetailer ~ and ended up with two posts to show the results 💖✨

The first post shows local without hires, local with hires, and on site ~ so you can see the direct comparison. The second one is another test without hires/adetailer. 💖✨

https://civitai.com/posts/22081762

https://civitai.com/posts/22082447

For full-body images, though, I’d recommend using adetailer locally and/or hires, or increasing the resolution ~ since faces can look softer or blurrier at longer distances, and this helps keep the output sharp and consistent. ✨

I’m tinkering a lot to make adetailer more optional by using one of my models but since I’ve only been doing AI for about a year I wouldn’t call myself an expert yet 😅💕

Maybe if someone else here uses Comfy and sees this, they could help out with that side of things, since I don’t really know Comfy myself 💕

@Ellie2024 I would love to help you, but I'm not familiar with comfy at all :I

Basically, I wouldn't say Comfy or Forge would be better, Comfy in particular seems to give you more space to influence the image. But it looked so complex that I said ... later :D I'm still at that point :D

I honestly don't know, I usually only use Forge locally.

VAEs are included in LvL1-3. In this case, I am not using a second one. I suspect in this case the output could seem experimental

To be honest, I do not know why this would happen. for me it works as shown in the showcases, either locally or onsite.

maybe the sampler?

hey ☺
thank you for your comment.
I'm glad if you like evo.
which version do you mean?
I usually don't use any tags for the hands at all in the latest version. Alternatively, I only write nails if I would like to bring them more into focus

@GasparGames Thank you very much for letting me know ☺ I'm very happy if it works now ♥

Have you ever tried 1boy, (adult) man, nude and then a movement/emotion? So is he sitting? leg bent, hand on the knee or hand on the back of the head, is he laughing? hairstyle, eye color?

it's important to write down what you want to see. To get the desired output, it is the input that needs to be described - referring to both facial expression and movement

hey, thanks for ur comment. which version are we talking about? :)

This cannot be completely ruled out. If you find a model that can do this 100 percent, I would recommend using it - if this is the most important fact^^

my art and models breath in the flaws ~ for me its part of the creative process :)

Basically, I can recommend the following when depicting hands:

- test models raw (before adding loras, quality tags, etc)

- If the image works as a blank, add everything you imagine step by step

- Pay attention to wording about hands, i.e. choose tags carefully (refers to hand motion, if there are too many different ones this can lead to a hand wanting to make more of a movement), tags like "good hands" can help, but can also have the opposite effect. If I want to consciously depict hands, I usually leave it out completely, at most I write nails if I want more focus here.

- please pay attention to denoising when hiring (the higher the number, the higher the probability that fingers and toes will be added because the pixels are stretched - especially for models that are intended for 1024x1024 resolution, 0.4 can be far too much)

- by adding many Loras in the same image: please step by step depending on the strength - everything in one can mean metaphorically depending on the image: I paint with oil paints, chalk, pencil, watercolor at the same time, add confetti and individual geometric cardboard cutouts - can result in a masterpiece, but that remains a matter of taste ^^

@CyberCurves thank u for ur comment, that might help but belonging to the enbedding and model it can have the opposite effect.

hey, what do u mean by running it through sdxl? u mean using sdxl lora? just asking cause its an IL model ☺

thank you both for your comment.

Currently only a limited number of models can be used for on site generations.

one of the possible spaces can be purchased in the auction house for a week at a time.

I have noticed, prices vary depending on the number of possible models.

Bidding on all of them would be unfair to other creators, in my opinion. I usually bid on a small number of models.

Besides, it would be too expensive for me too, to be honest^^

♥ Reij

@dawushi hey, I hope the problem is fixed? :)

I wanted to point out that the model contains a VAE. So please don't use any additional ones :) then it should work.

thank you very much for your kindness ♥ I am glad if you like EVO

hey, thank you for your kind feedback :) to be honest, I actually already thought about it today. But I can't say exactly when that will be