So basically, it's a text improvement LoRA, but it also acts as an aesthetics enhancement LoRA (which was unintended, but still welcome). Beta version.
How to use:
There are three main trigger words: "speech bubble", "text", and "snapchat".
While the first two are self-explanatory, "snapchat" adds a Snapchat-like semi-transparent bar with text on it.
The basic pattern is: {trigger word} "words words words".
You can also use [n] in some cases to create a speech bubble or a string of text (Unstable!).
Just look at the prompts in the example images :3
You can also use this lora without any trigger words, just for an aesthetics enhancements.
Train info:
15 epochs, 2925 steps
Dataset: ~400 imgs, manually captioned text, WD3 Large for tags
Batch size 2, gradient acc 4, keep tags 5, shuffle the rest, no dropout, TE was NOT trained.
Resolutions = [768, 1024, 1280, 1536]
Trained on WSL2, RTX 5060 Ti 16gb for ~16 hrs.
I want to continue working on this LoRA; however, captioning the text by hand is a huge pain in the ass.
Description
Doubled the dataset's size, switched to LoHA, tried to train LLMAdapter alongside. Trained on Base-v1. Stronger aesthetics? Don't expect much.
FAQ
Comments (5)
This is the only Lora I found that do text right. I'm in love.
Any particular reason it was made for Anima?
Thanks!
Initially I was experimenting on SDXL checkpoints, Illustrious specifically. It was a complete failure though, mostly because the text encoder (dual CLIP) was just too weak.
Then Lumina, NetaLumina, and NewBie Image came out one after another. Strong and effective for their time, but still too heavy for my setup (rtx 3060ti 8gb) to train stably.
Anima ended up being the sweet spot between weak-but-outdated SDXL and strong-but-too-heavy models like Flux, Krea 2, or Qwen Image. Basically SDXL 2.0 for me.
While this can help with text on clothing would it be possible to get a trigger word specific for that to get a little more consistency?
I'm actually planning to rework the caption format again, so it accounts for the text specifics on the image, that might give more consistency than a dedicated trigger word alone. Hoping to have a proper, non-experimental version out in a week or two.
would it be possible to use bubble without pointing or pointing to corner of pic. I prompt it to point at corner and it didn't work
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.