Usage
Trigger: text logo
Weight: 0.5 - 0.8
As always experiment with these values.
Recommended model: ZavyChroma v5
Recommended prompt format, although other styles and keyword positions seem to work too:
("YOUR TEXT" text logo:1.2), some description of what happens, simple background <lora:dvr-ftl:0.8>
Well, this has been a journey ... this is version 14 and it's not bad but there's room for improvement. As always check the lora gallery and showcase for examples.
Warning:
Some of the showcase examples have been re-rolled a few times to get the text to work. Text in SDXL is still a pain, join the SD3 waiting list. Some words and word combinations will work on your first attempt, some will not work at all.
I couldn't get this to work reliably with Juggernaut v9 Lightning with the standard recommended sampler DPM++ SDE Karras . I found Euler SGMUniform to be marginally better but not good enough to produce consistent results.
Description
FAQ
Comments (24)
text generation in sdxl is making me go nuts. It simply just takes too much time to get right..
that's true, let's hope SD3 makes things better
Harrlogos has competition?
a friendly rivalry until SD3 comes out
@Dever what if SD3 has bugged text too?
@yofoton174609 the examples they shared so far look promising so let's see what SD3 is capable of first
@Dever yeah but they shared text samples for stable cascade too, yet stable cascade fails to make text most of the time
Celebrating 1k downloads 🥳 Thank you all, I hope you're having fun.
This is so good. Having a lot of fun. Good job.
Thanks, glad you like it. Share what you make if it's shareable. Leave a review, like and subscribe and all that :)
Outside of being a text ia this is really a great logo creator, the propositions really give incredible vibes
what resolution i should use?
standard SDXL resolutions should work, if in doubt you can use 1024 x 1024
@Dever using ("RAT KING" text logo:1.2), king rat with his rats, advert, realistic, castle background <lora:dvr-ftl:0.8> this prompt + 1024 x 1024, dpm 3m sde, 30 steps and it dont work at all for me ;/ letters are completely mispelled
@Polack90pol unfortunately that is what we're working with, the base model is still not great at text. The lora helps a bit and guides towards a logo. SD3 is much more consistent but even that will not get it right all the time
@Dever Thank you for info, do you have any suggestion how can I improve the result? Maybe chenging sampler, or cfg scale
@Polack90pol best thing you can do is try multiple seeds, I'll have a go too and see if I can create one for you
@Polack90pol here https://civitai.com/images/16369884
@Dever i see that you used different checkpoint, is it better?
@Polack90pol I just used it because it's an SDXL merge I made but you can use whatever checkpoint you want, don't use Pony as the lora is not compatible with Pony
@Dever im using zavychroma v8 but idk
Good job
i dont know why but i do not get same results
your models are doing some heavy lifting with the base with a few others... i am trying to dial in a easy setup switch "make your own prompt".. so people can get fun text images : )
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.