Texta is a mix of LoRAs I trained on pictures containing text, signs and logos.
Success is not guaranteed! But it should help improve your chances. 🙂
If you get close, keep the seed, change the number of steps or tweak the CFG to get a better result. There is no perfect setting, and will depend on which base model and other LoRAs are being used with this LoRA.
I have been using DPMPP 2M SDE Karras with CFG from 3 to 7 and steps from 20 to 100. If you are getting completely scrambled text, I found increasing the steps can help.
Description
FAQ
Comments (15)
If you load it while using a SD1.5 model it also has a kind of effect in this direction. Works better when using FreeU2, but of course still not the great result as in SDXL. Any other recommendations?
I get the following error, when trying to use it in SD 1.5
RuntimeError: The size of tensor a (768) must match the size of tensor b (2048) at non-singleton dimension 1
Time taken: 1.0 sec
Because it's SDXL
The author stipulates base model of SDXL 1.0 so why would you use SD 1.5?
my bad. I wasn't paying attention. I will leave this up in case someone makes the same foolish mistake.
i tried words like INVINCIBLE or POWERFUL and it works like sh*t. It always create some weird word with 2 or 3 repeating letters or weird spelling. i added the negative prompts and it still does this.
i think it might be from the checkpoint. i use basic fooocus on google colab and this text lora. i will try using the Checkpoint too. i just wish it uploads asap on google colab and i don't get these weird misspelled words/content
Long words are still hard to generate. Certain letter combinations can also be difficult, like V and W. It gets single and double letters mixed up. It would require a lot of training to fix all these issues. I trained the top 300ish most common syllables in the English language.
StableCascade makes the best text of all the models right now. but even StableCascade sucks.
@yofoton174609 its flux now. By the way, im from the future.
It works but the text is in Japanese how do I change it to English?
nani?
@humblemikey oh god its infected the comments!
I haven’t been getting any text at all
I was successful in generating legible text using the checkpoint JuggernautXL.safetensors
I get the text messed up...
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.