All samples were generated using my custom ComfyUI workflow linked down below and may not represent the quality you would get using a different workflow or the CivitAI online generation service. Samples are also slightly cherrypicked by having a tolerance of up to 3 seeds of lower quality generations (often anatomy) or generations not adhering to the prompt enough.

Changelog:

v2 [26.11.2024]

Much less overtraining and smaller size by changing the training config by going from 32/32 dim/alpha to 16/16, removing caption dropout, and adding a 1e-7 Clip L text encoder, reducing training image count from 30 to 15, and generating new captions by ChatGPT and editing them

Known issues:

Generations using the CivitAI generation service may be of lower quality, due to the simplicity of the CivitAI generator
Not every seed has accurate prompt following or is free of anatomical or other quality issues. Still, I found that generally a good generation should be guaranteed within 2 seeds.
A few rare prompts, often some modernity based ones, may not work well
Combination with other LoRa's, e.g. celebrity/character LoRa's, may not work very well (Sample size: 1)

Before submitting a bad review, please make sure that you also tested your prompt using my recommended inference workflow, once with the LoRa applied and once without, on four different seeds. Often bad images result from deviating from the settings recommended by me or are actually the result of a fault within the base model that I am not responsible for.

LoRa description:

This LoRa model was designed to generate images that look similar to the artstyle used in the Darkest Dungeon game.

Trigger words:

inked comic illustration with flat colors shading thick outlines in dd artstyle picturing

See also the prompts of my uploaded sample images for how I did it!

Recommended inference workflow:

1.0/1.0 LoRa model/clip strength
deis/ddim_uniform
20 steps
3.5 FLUX Guidance
1 CFG
No negative prompt

Link to my personal ComfyUI workflow using the above settings:

https://www.dropbox.com/scl/fi/7m664pgslavb5uoobovmm/dd-workflow.json?rlkey=z1r7juh6ogsn04ys4ewndlupd&st=viiykxj5&dl=1

Tested on the flux1-dev-Q8_0.gguf base model with the t5-v1_1-xxl-encoder-Q8_0.gguf and clip_l.safetensors text encoders, using a RTX 3070 8gb at ~1min 30s generation time for both passes.

Training Workflow:

15 images with captions generated by ChatGPT and edited by me and trained for ~1h 15min on the consolidated_s6700_fp8.safetensors base model by MinusZoneAI and t5xxl_fp8_e4m3fn.safetensors using the GUI version by bmaltais of the 'sd-scripts' trainer by kohya-ss on a RTX 4090 24gb GPU rented on Vast.AI for ~0.45$/h.

My training configuration .json file with all the settings I used for training this LoRa:

https://www.dropbox.com/scl/fi/nmv23p5ldwakous8p79iu/FLUX.1-dev-LoRa-Style-Darkest-Dungeon-v2.0-26.11.2024-SPECTRUM-0002-by-AI_Characters-_20241126-181533.json?rlkey=fi0ayq3cw059engvtctlxp7ym&st=vib03h44&dl=1

Disclaimer:

I do not claim to be the current owner, original creator of, person or legal representative of a person depicted in, any of the concepts that my models, including this one, aim to emulate nor the training data used to train this model. I do not claim that any of my models have received an endorsement by those respective individuals. I also do not claim that the emulation attempted by my models represents a 100% accurate depiction of the original concept in question nor do I claim that the quality of my emulation reaches the same quality of the original concept. All credit goes to the respective current owners, original creators, or people and I encourage you to support them in any way you can.

If you are the current owner, original creator of, person or legal representative of a person depicted in, any of the concepts that any of my models aims to emulate and would like that specific model to be removed, then please leave me a private message here or on Reddit with proof of authenticity of your identity and claim and I will remove it.
Additionally, I do not endorse my models being used to violate the law or enact still legal, but immoral acts such as the spread of misinformation using deep fakes.

Donations

All my public work will forever be free to use. Likes, downloads, and good reviews are already support enough.

However, if you would still like to support me financially, please do not donate Buzz to me, as I have no use for it other than promoting my work on the front page of CivitAI. If you have money to spare and want to support me, it is better to donate to my Ko-Fi: https://ko-fi.com/aicharacters

All money donated there will go towards covering my ~5000€ costs accrued over an entire year of me renting multiple RTX 4090 24gb and A100 80gb GPU's almost every day to train over ~1000 variations of test models with different underlying base models, datasets, and settings for SDXL, FLUX.1 [dev] and SD 3.5 Large in order to find the most optimal combination of base model, dataset, and settings.

Commissions

I currently do not offer commissions, however I may soon in the future depending on my primary place of work cooperating with me having a side job as well as there being enough interest here in me offering commissions.

Commisions of real people other than public and well known celebrities or influencers (e.g. yourself, coworkers, family members, friends, some influencers) require a proof of authenticity that cannot be easily faked using AI and this proof must come from the person you want depicted. I also reserve the right to decline any such commission for any arbitrary reason that I may have, regardless of proof of authenticity. Commissions of real people that cannot give consent such as children or mentally ill people, will lead to your commission being ignored. Dead people will be decided on a case by case basis.

Commissions of NSFW and political concepts will be decided on a case by case basis by me.

Any commissions of outright illegal concepts will be ignored and your account blocked and reported. Any commission of a still legal, but morally questionable act such as deep fakes to spread misinformation, will lead to your commission bekng ignored and your account being blocked.

I also reserve the right to publish under my account any models you commissioned for a public concept (e.g. a popular artstyle, character, basic concept, celebrity, etc … that you are not the current owner or original creator of or that is not something of a private nature) that other people may also be interested in using.

The commissioned models would also never deviate from my current training workflow, e.g. settings and dataset routine would be set by me to the same ones that I use for my own work. You would also have to supply me a dataset of the 30 highest quality images you can find or pay an extra fee for me to gather and edit that dataset myself.

Of course, all of this may be subject to change in the future.
If after reading and agreeing to all of this you are still interested in commissioning from me, then please fill out this little strawpoll so that I know how much potential interest there is and in what price range: https://strawpoll.com/jVyG21EvYZ7

Search AI Models & Addons

very few results