- trained on modern logo's from interest
- use "abstract", "sharp", "text", "letter x", "rounded", "_colour_ text", "shape", to modify the look of the logo
- use "modernlogo" at the beggining of the prompt
Description
FAQ
Comments (14)
its "Logos" not "Logo's"
Logos is plural for Logo, as in more than one Logo
Logo's is possessive, as is something belongs to a Logo.
That means you can hit the generate button again 👍 your welcome (I know it's "you're")
@kleind Touche, my friend :D
Also missed the "ooff"
sry. just piling on here :)
Great! I'm curious about which checkpoint model you used. As far as I know, LoRA files should be used with a checkpoint model. Can I just apply LoRA to the default checkpoint model?
How many logos did you use and which ones?
Does it generate with an empty background?
Running on 4090
Getting this error
should this go inside lora folder or stable diffusion folder?
NansException: A tensor with all NaNs was produced in Unet. This could be either because there's not enough precision to represent the picture, or because your video card does not support half type. Try setting the "Upcast cross attention layer to float32" option in Settings > Stable Diffusion or using the --no-half commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.
Time taken: 0.31s
Torch active/reserved: 2249/2376 MiB, Sys VRAM: 5012/24564 MiB (20.4%)
No you need to go to Settings Tab an then on the left side select stable diffusion then scroll down and activate "Upcast cross attention layer to float32" then apply settings and reload ui then try again
May I ask which model you are using
is there any prompt to have a specific wording logo?
SD is shite for reproducing text RN, Best you can do for now is do img2img on some text, or use an image as a depth or canny controlmap and use controlnet, then fiddle around with the weighting of everything so that you can let it slowly creep towards a different style. Its not a great workflow though, the controlnet stuff is better for making like high detail renders of your text in weird styles, rather than changing the shape of the text itself.
How to fix letters
that's hopeless, stable diffusion 1.5-based models can't do lettering 99% of the time. You might get lucky if it is a common word or 1-3 letters, if it's in English. This is just good for getting a general idea and cleaning it up in photoshop
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.