The latest version workflow and nodes are here
https://github.com/peterkickasspeter-civit/ComfyUI-ZImageTurboProgressiveLockedUpscale/tree/experiment
=====
Custom node: https://github.com/peterkickasspeter-civit/ComfyUI-ZImageTurboProgressiveLockedUpscale
Workflow is in the showcase images. More details here - https://www.reddit.com/r/StableDiffusion/comments/1r1qyj2/zimageturboprogressivelockedupscale_works_with_z/
Distill lora for base - https://huggingface.co/alibaba-pai/Z-Image-Fun-Lora-Distill/tree/main
*use either Z-Image-Fun-Lora-Distill-8-Steps.safetensors or Z-Image-Fun-Lora-Distill-4-Steps-2602-ComfyUI.safetensors
Description
Qwen
FAQ
Comments (47)
Welcome Back!
The Return of The King!!!, i still use your flux amateur photography lora till this day aint no lora beating that
oh wow hahahaha. thanks man
I tested it with Q5 Qwen model and the workflow you used. the output is just noise any idea why? https://files.catbox.moe/fxre6w.png
This lora is amazing! Only bad thing is it gives huge tits to any girls, even clothed & casual. If planning a v2, maybe add some chest size variety in the training so not all the women have huge ones
oh, interesting. you can reduce the lora weight and try to tweak the prompt. i'll see what i can do
all women have huge tits inside but awaiting for the proper time to popup ..
@amazingbeauty did you try popping them out?
@peterkickasspeter not yet but i am trying to do asap
Well, the ONE THING I thought I'd never hear anybody complain about on the internet.... welp, here we are.
The best LoRA I’ve used in the last 4 years. Great job
What are you talking about, what 4 years? 😅😅
Bro this looks so good, a lot of times, for some reason, lora makers think amateur photography means photos that look 0.5MP from 2005, thankfully this one doesn't have this problem. Thanks.
Bro honestly your work is absoulut amazing. I used your Flux Lora and yet i use your Qwen Lora and both are the best Loras i ever used how you get this Quality thats crazy...
What is the effect of this LoRa?
is "photo in the style of redditya" that the trigger? or why do most examples have that phrase? what different styles does it understand?
trained the lora using that phrase. you can use it. sometimes it makes a difference, sometimes it doesn't
also it tends to make that harsh flash-camera effect ..how can avoid this ? maintain realistic look but with normal light or prompt can control the light ? that camera-flash is intense how to avoid it ? or maybe i use it wrong ?
I hate to sound like a butthole here, but harsh bad lighting is the number one staple of amateur photography.
Could try relighting it with that qwen image edit lora, might be on civit but i got it from a huggingface repo.
Amateur photography's defining characteristic is crappy lighting and candid framing. If you want realism I suggest finding or creating a realism lora that focuses on professional photography as your dataset.
@heylelheosphoros666 when i have a gpu.
Testing qwen,most of the time the characters are holding a paper with words on it,it happens a lot in pov view
training bias. detailed prompts help avoid this but i am planning to fix it in the next version
@peterkickasspeter awesome, btw ,are you using qwen base or other checkpoints?
@Neon_signs base qwen. I don't use finetunes, they introduce unintended biases
The upscaler seems to suck away some of the skin texture.
I think so too. If the image I'm feeding to the upscaler is not sharp, it degrades the quality. In this case we need to downscale the image even further to add details but it's really bad for text
Hey man, what datasets did you use to make this lora? Like some realistic images?
Yes correct
@peterkickasspeter Can you tell how many images you used? Plus what type of images you used? Like nature photos, characters, cars, etc?
@coolstrad Mann goodluck people always ghost when you need them ahaha. Try to go on lmarea and ask gemini. Lemme know when you find the answer
@Thatonecreator Yea it’s just sad. I don’t get it what’s so harmful to share the dataset details.
@peterkickasspeter 👍
Please please run your dataset through chroma!
Sorry, I'm not willing to spend 50$ on a model that I'll never use
@peterkickasspeter why 50$? do it through civitai, its hardly 5 bucks?
@bhass oh, chroma lora training in available in civitai?
@peterkickasspeter yes!!
@peterkickasspeter how much buzz you need to train for chroma?
@doorajar24 i am not exactly sure. i stopped using civitai lora trainer a long time ago. i have to check
minimum of 1.3k buzz for chroma lora training I think.
Why Chroma?
@runebloodstone chroma was suppose to be the next level of pony/illu that's why, but things doesn't go that well
the GB size is too big for most services to run this on like FAL/Comfyu Online, do you have a shorten version?
Damn bro the preview pics looks amazing, it completely removed that annoying ai feel from the pictures, i am gonna test it right now. Thanks!
Details
Files
amateurphotography_v1_qwen.safetensors
Mirrors
amateurphotography_v1_qwen.safetensors
Qwen_LoRA_Amateur_Photo_v1.safetensors
amateurphotography_v1_qwen.safetensors
QWEN_AmateurPhotography.safetensors
amateurphotography_v1_qwen.safetensors
amateurphotography_v1_qwen.safetensors
amateurphotography_v1_qwen.safetensors
amateurphotography_v1_qwen (1).safetensors
amateurphotography_v1_qwen.safetensors
amateurphotography_v1_qwen.safetensors
amateurphotography_v1_qwen.safetensors
amateurphotography_v1_qwen.safetensors
QWEN_AmateurPhotography.safetensors
amateurphotography_v1_qwen.safetensors
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.