CoffeeMix is intended primarily for producing more cartoony, flatter anime pictures that tend to have more pronounced lineart and cel shading.
This model is not intended for detailed illustrations so backgrounds will be on the simpler/cartoony side. It works best with simple, short prompts and I highly encourage trying fewer tokens(prompt length and embeddings) if you dislike the results.
Use EasyNegative on negative prompt. VAE is baked-in.
My recommended sampler is DPM++ 2M Karras with 20 sampling steps, but other samplers work just fine. Model is pretty sfw unless you add nsfw into positive prompt.
If you like the model, please consider tipping me a coffee.
Description
v1.5: More definition in lineart, cel shading. LoRAs work better and eyes are more detailed.
FAQ
Comments (5)
do you aslo upload it on Hugging Face?
Just made a repo on huggingface, it's available there now.
CoffeeCoffee/CoffeeMix-v1.5
A model that can output enough for a short prompt is a good one. Good job.
Question to the author, what do you think, should this model be "flatter picture" like "flat_illustration" or "digital art of"?
I'm in the "more flat is fine (since it's a model that aims to be flat), recommended faction". I think the addition of the prompt makes it more tasteful.
Do you mean in the prompt? You don't need to specify flat colors since it's biased to produce flatter images. Typically describing a color will make it pop out in the output and be saturated.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.