How to Use
This is a style LoRA, designed to be the base layer of your LoRA stack. It creates the foundational aesthetic of realism, upon which you can add character or concept LoRAs.
Note: The included ZIP archive contains both the high-noise and low-noise LoRA variants, along with our recommended ComfyUI workflows.
Trigger Word:
InstacamRecommended Strength:
1.0. Start here and adjust in small increments.
I would also like to thank Danrisi, who originally taught us how to train LoRAs and helped make our work possible.
Description
Instagirl v2.5 is a significant update focused on increasing model diversity and detail fidelity.
Expanded Dataset: This version is trained on a new, highly-curated dataset of over 2,400 images. The primary goal of this expansion was to reduce the "signature face" effect of previous versions and increase the variety of generated characters.
Increased Network Rank: Instagirl v2.5 low noise model was trained at a network rank of 256. This provides the LoRA with a significantly greater capacity for learning fine details, which can result in improved skin textures and more nuanced lighting, particularly in close-up compositions. High noise we have kept as rank 32 for smaller download size as it does not benefit from the higher rank which only helps with details.
FAQ
Comments (34)
There is only one lora file. Was it trained on the high noise or low noise model?
The download is a .zip file. Make sure to dl it as .zip and unzip it first :)
@VLRevolution Word. I didn't notice that. Thanks!
is there a lowvram version? im working with RTX3060 12gb ram, i hope run this model.
2.5GB low??? v.2.5
This is supposed to be for wan 2.2 and all they post is pictures?
Wan also does images.Not a video only model
I don't trust ziped archives, please, consider publishing only the safetensors
.zip also makes it so I cannot import directly to MAGE and other sites
It's not possible as we need to include both the high and low noise in the zip. I'm sure with already almost 20k downloads you should be able to trust us.
@Instara It would be better if you publish the two of them separated because some of us need to use it on pages like Runpod—I'm pretty sure it's the most common case since almost nobody is going to wait 6 minutes per video—because now it's not like we can just download it and use it. And, come on, not everyone do know how to do so and adding extra steps bothers a lot.
Is there any advantage on having the two of them in a zip file? That's weird. Why would you put it in a single zip file instead of doing like everyone does which is the most accessible and simple way to publish them? I'd understand the concerns of the original comment.
@Instara Nearly every other lora is published with separate high and low, why only you need the be the different one?
I mean, you didnt provided any convincing argument about that. Which makes this even less thrust worhy. Since civitai don't supervise the contant, is just fair we would get just the safe tensor as everybody does.
20k gooner dumbasses downloading it not knowing the risks, doesn't make it thrust worthy, i think civitai should make this kind of file sharing not possible.
We literally have 25 GB models as safe tensor but yours need to be as zip?
Cmon bro, just see as we, it's really not that hard to understand
I really do appreciate you even making it available, but put our shoes for one sec, it's really not that hard.
Most people here produce nsfw content.
It's just fair we know what we are downloading
@Instara You're full of shit and you know it
lol everything nis on bid lol this site now its scam
is this only for image or also for video?
You set the # of frames to 1 to get a single image.
The LOW-NOISE is 2.3 GB and the HIGH-NOISE only 292 MB, is it correct? 🤔
Correct :) check the change notes
Getting error with this workflow: KSamplerAdvanced
UnboundLocalError: local variable 'cl_path' referenced before assignment
this model is awesome, can you please add more backshots on dataset? Like add some "in all fours" images, model tends to sit either on the floor or bed, expand the dataset to more pictures of the back of the models if makes any sense.
Thanks, im enjoying it a lot.
Please re-upload, separate the high and low noise models for uploading on different pages, do not upload as a zip. Thank you. The current method cannot be downloaded in one click by the LoRA Manager.
You can upload the individual files to your own storage solution. Then you can download the files using rclone from command line.
@MrSjeikz Too much of a pain, give the version id's so that we can auto download them via API when spinning up a cloud instance.
Please upload as 2 separate files for high and low as 99.9% of other people do. Some of use use LoRa Manager.
I keep getting color shifting at the beginning of videos when using v2.5 but not v2.3, anyone else?
does anyone have a workflow for img2img edit? I heard it's possible and want to try it
dm me
dose these models work for sex and nude isn't there any think else can do with it
should I raise the loras "up" to get more detailed back grounds?
does anyone have a workflow for img to img?
Muy bueno, muchas gracias.
does anyone have solution to use this flow in mac? seems like it only supports fp16 instead of fp8 :(
what is your workflow supposed to be