UNDERGROUND Comics FLUX
Underground comics now for Flux! With detailed, cartoonish characters from comic book covers of the 60s-70s
Make Covers:vintage comic book cover
Make Artworks:vintage comic book illustration
Make Comic Strip Panels :Two vintage comic book panels featuring […]. In the top panel […]. In the bottom panel […]. "
Guidance:5 - 8
––––––––––––––––––––––––––––––––––
LEGACY Edition FLUX & SD3.5L
Remaster of the original SDXL version. Using the same dataset and more or less the same recipe.
Capable of making both covers, as well as artworks:
Cover Trigger:vintage comic book cover
Artwork Trigger:vintage comic book style or simply avoid using quotation-marks or mentioning title/text etc
Remember that Flux & SD3.5 prompts are case sensitive, this is especially useful for titles.
See 'About this version' for more info.
Support your Wizard – post your images to the gallery ♥
––––––––––––––––––––––––––––––––––
FLUX Sci-fi Pulp
Make your own comic book front page.
The LoRA adds better ’grittyness’, ’vintageness’ and ’pulpyness’ compared to base flux. Trained on 23 pulp style comic book covers from the 40s-50s.
This LoRA likes sci-fi stuff, so feel free to embrace the retro-futuristic aesthetics by using word like: Ray gun, Glass helmet, futuristic, rocket ship, science fiction etcetera
If you're not getting the pulp aesthetics right, try prompting 1950s pulp art, pulp novel style, or similar.
’Small callout burst’ is great to use. You can also get more vintage feeling with ’creases’, ’used condition’, ’damaged corner’, ’stains’, but it can easily go overboard.
Happy to see you share your creations in the gallery below! Any buzz-tip goes straight to finetuning more Flux-LoRAs
––––––––––––––––––––––––––––––––––
SDXL
Check out the version tab for different styles / eras.
Latest version is grittier, less pulp and more focused on ink. Inspired by the underground comics of the 50s - 70s and artists like Will Eisner, Robert Crumb and Rand Holmes.
Hopefully you'll notice a better text coherace and better variations on the titles.
Use a 2:3 aspect ratio and keep your prompts simple.
To get better titles use this format: vintage comic book with the title "xyz"
Try different checkpoint models for more or less 'illustrated' look. I recommend PixelWave or PixelWaveTurbo for more illustrated look, and TurboVision for a more detailed and semi-realistic type of illustrations.
––––––––––––––––––––––––––––––––––
SD1.5
• Use a 2:3 aspect ratio and keep your prompts short and simple.
• Hi-Res fix and 768x1152 is recommended
• Showcase images use Dreamshaper
Description
Remaster of the original SDXL version, here is the Legacy version for SD3.5L.
I have mostly used these settings:
LoRA Strength: 0.8-1.0
Sampler: DDIM SGM_Uniform
Guidance: 3.5-7.0
Model: SD3.5L Turbo
Steps: 8
Size: 832x1216
Timestep Shift: 4 or 6
Trained on a single image, so naturally it is overfit in some areas; mainly typography, pose and style, (the last one being deliberate). Will release a version with bigger a dataset in the future. Get more variation/flexibility by lowering the LoRA strength
FAQ
Comments (6)
Hi. What are your first impressions of SD3.5 esp. with regards to your Loras - same, better, quicker etc than Flux?
Speed-wise: It's like changing from clogs to running shoes.
Training-wise: It's like wandering off the marked trail, where you could end up either knee-deep in a bog – or find a beautiful, secret waterfall where stunning valkyries are skinny-dipping (they all might have six-fingers and a cleft chin tho)
I love that the gap between SD3.5L-Turbo / SD3.5L is smaller, than the gap between Dev/Schnell or Dev with any acceleration LoRAs. So much easier to test different epochs.
It feels like SD3.5L gets much closer to the training data – Flux is kind of like an Auto Complete-option that you can't turn off: Yes this looks nice, but it wasn't really what I was aiming for.
That's my first impressions at least. I'm glad Stability AI had the big backlash and is back in the game, otherwise it would be a race towards closed source and higher safety walls. Only concern I have is if the community gets spread out too thin.
Have you tried it? What are your impressions?
@WizardWhitebeard Thanks for the review. Much appreciated. No I haven't had time to try it yet but looking forward to it. I'm glad Stability AI are back too - a bit of competition helps. Yes hope the community doesn't spread too far and contributes to SD 3.5 too - early days but it looks like that is happening.
Big fan. Keep up the great work!
thanks for it
cool. This is the first SD3.5 Lora i see :D
You always make 10/10 stuff!!
Don't know which version to run?
The pinned posts, in the Legacy SD3.5L-version gallery, are comparisons for the different versions of the LoRA.
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.