Give a bath of oil to your QWEN Images and make them look shiny and glossy! If you like my work and want to support me you can do so by donating Buzz or by PayPal https://www.paypal.com/donate/?hosted_button_id=L87JAGTHDNBRA
Note: Please if you see any of my models somewhere else report or let me know! Thnx
Support creators:
by liking their models (many forget to do it)
by publishing your photos directly in the model's gallery.
by getting involved in their Early Access programs
Thank you for your support.
Description
FAQ
Comments (10)
Since Flux have a tendency to give glossy skin by default, a Lora for Dry Skin would be useful.
well one thing is having a partially kinda glossing skin than an entirely body with the control of how much glossy you want with the lora. Flux by it self don't offer control that's why fine-tuning is need. Also Flux make a bit glossy but not enough to simulate oil on skin in a more realistic way but thanks for the feedback and I will consider in making a dry skin lora as you suggested.
@Artikuz_Ai
I did mean your lora is it useful, and I will use it. Sorry if my comment gave you the wrong impression.
@strong134 is ok, and it dint bother me I understand than some people can see things differently in a new or different perspective or angle, like you mention a dry skin lora idea and that never occurred to me and thanks to your feedback now I have something new to work with and offer it here on the platform. Once again thanks :)
@strong134 heheh, it made me chuckle because I'm still waiting for the loras that will allow Qwen to do skin and lighting as well as Pony.
@duAIguy so you prefer the realistic skin from pony? Im a bit lost because as a photographer I like more the QWEN but mostly Flux Krea they look more realistic and natural but yeah in some occasions I will use the posy aesthetic.
@Artikuz_Ai Qwen looks like basic Flux with no lora, or like using Wan for t2i. It's not photographic at all. It renders people as plastic. Krea, with the right prompt, can make something pretty convincing (and often more dynamic than the others). Krea's problem is it doesn't always play nice with other flux loras and so getting it to follow a prompt, sometimes even a simple one, is difficult and frustrating. Just like Flux itself which often needs the increased difficulty of kontext if flux just doesn't know how to do something simple...like sit a character on someone's lap, for instance, without growing false limbs or forgetting about others, or the spatial relationships between people and objects.
Pony on its own isn't good but there's loras that let it do more photographic skin than any image from Qwen you might care to reference. Take that as a challenge if you like but Qwen, at this point, is like looking at AI people from the past. People talk a lot about Flux chin, well, Qwen has a ways to go before it shakes off Qwen skin and Qwen lighting.
It's especially bad since the speed-up loras kill detail and quality, in hair, in materials. Not sure yet if it suffers more than others from smaller quants but little 6gb Pony fine tune checkpoint with the right loras can shit on Qwen...it's just also likely to give you mutants.
It's just funny that your lora here puts a fairly realistic cover-up on a basic sin/failing in Qwen on its own in the same scenario without your lora.
@duAIguy well yeah in my four years plus studying and working with ai as a professional as a free lancer I have tested a lot of ai models and have seen that their good at some stuff and excels at other. And I I agree I have seen bad qwen images almost all decent and very few that let me with my eyes wide open contemplating all the details as a realistic style fan I am jeje. flux and chroma was pretty good but like you say u ask for a pose and u get a pony lol and want thing I also love about illustrious and pony most likely noobai too the are so good following our desire to make them true. QWEN is more like for marketing stuff in my opinion but is the new guy in the park flexing and showing off probably just to try to intimidate the other models jeje. I kinda love krea but it tends to add to much noise in occasion also to overexposure the images so we need to fine-tune the prompting. what I like to do is use my favorite tool, my dear comfyui to make an image let say with a pony model a dmd2 is possible as a draft for a speed up a concept and if I like it then I recreate it with one of my other favorite models for the style or effect im looking for. by the way what is your workflow? if any do you work locally or here on civitai? Im preparing a dataset to see if I manage to create a lora with a more realistic skin effect for pony and illustrious, I have try some but they dont look very realistic to me but they are decent for sure.
@Artikuz_Ai I only work local and will never use a cloud service
I looked at a lot of different WFs and saw what worked, what didn't and what was curiously present in some while absent in others without that WF being totally bad or necessarily even worse than any other. But for image gen it's a basic two sampler with latent upscale (because upscale models on final images all introduce distortion and artifacts that make it worse, like the garbage "4K" blurays coming to market). And then a face detailer is almost always needed.
Sometimes the latent upscale fixes bad hands. Moreso with a model like Flux but not always. It never seems to fix bad hands in Pony, just makes any deformity sharper, hahah.
I played around with hand detailers a bit but they aren't a panacea and don't seem to work as well as face detailers, at least not the ones I tried, so I'll generally just re-seed and try again. But I'm also not making anything just for imagery it's to feed i2v. I feel like the best and most convincing stuff is almost always starting from a good inference image setting the aesthetic before you do the longer gen.
@duAIguy yeah i agree, I like always to make a few images two choose at least 3 as finals sometimes use one for a vid. I love to play and experiment with samplers and schedulers but like you say those models bare offer 1 or 2. also I have been noticing not all platform have same settings because some of my loras work best in some and other not no much.soon we will have nano banana here on generation but at the moment i think i will stay with krea for a little while for realistic at least. i hope they bring Qwen image edit soon I want to test that one too.