Sign In

Instagirl WAN 2.2/WAN 2.1

169

1.9k

10

71

Verified:

SafeTensor

Type

LoRA

Stats

381

3

Reviews

Published

Aug 1, 2025

Base Model

Wan Video 14B t2v

Hash

AutoV2
9CB7DEDDA1
default creator card background decoration
00quebec's Avatar

00quebec

CLICK "ABOUT THIS VERSION"

TRIGGER: Instagirl

Danrisi was such a huge help to me and has been for the past few months and I probobally wouldn't have been able to get this LORA so good without him.

I'm honestly really proud of this, and this is only my first iteration. I plan on further bettering the dataset and eventually fine tuning WAN 2.2 with 3000+ images and 130+ different instagram models, so this was really just a test run. This model was NOT trained on ANY NSFW, but if your really a gooner like that I can reccomend Flux Kontext (also maybe it would be good to fix text in images)

Generation time 1024 x 1440: 26 sec on h100, 58-66 sec on RTX 5090

We are currently working on speeding it up, but feel free to DM me if you have any suggestions. Discord linked in description.

I reccomend using Danrisi's WAN lora at around 0.5-0.3 strength:

https://civitai.com/models/1662740/i-dunno-how-to-call-this-lora-ultrareal

Here is my ChatGPT system prompt I use, how it works is that I will find an image I want to replicate on pintrest/instagram and paste it into ChatGPT with this system prompt:

Write one run-on paragraph (about 70 words) of concise, comma-separated phrases with no quotation marks that recreates every attached reference image for Flux.

• Start with the subject and LoRA trigger exactly: Instagirl, l3n0v0, no makeup, petite body, then continue in this order: pose → camera angle → person & ethnicity description - clothing & key accessories → environment/background → lighting → extra atmosphere.

• Keep the phrasing natural but brief—think quick notes to a friend—each detail separated by commas rather than full sentences (e.g. new york city, gloomy day, beautiful girl, taking a selfie).

• State that the photo is amateur cellphone quality and make sure to clearly state the hair color, type, and facial structure; if and only if it’s a mirror selfie, add shot with a silver iPhone (three cameras) held by the subject.

• Ignore all tattoos, piercings, body mods, glasses, on-screen GUIs, or icons.

• Avoid filler words and avoid synonyms for the LoRA triggers so their influence stays strong.

• After the paragraph, on a new line, list exactly: visible sensor noise, artificial over-sharpening, heavy HDR glow, amateur photo, blown-out highlights, crushed shadows.

If there are no subjects, ignore the subject part completely