Bigblumix | Artists Styles Checkpoint | PonyXL

Hi there and welcome to another checkpoint using my finetune based on PonyV6, aimed at having 230+ working artist styles without going crazy or training over existing knowledge ( for example fumihiko and hews) with some custom tokens so the styles can get trained from scratch, some styles are a bit underbaked due to stuff, but you can raise the weight and they can work, wasn't able to generate samples for all artists styles but all triggers in the txt file below do work tee-hee-pero~~

Please use the style triggers to change how the checkpoint looks, you can gen with or without Score tags, and up to cfg 8!

Two versions: v1 which is Bigblumix-BaseFinetune +AutismConfetti at 0.3, and "Bigblumix-BaseFinetune", which is the finetune as it came out of the bakery, you'll get thiccer girls by default as the majority of artist styles in this finetune lean that way, in comparison Bluvoll's 3px is AutismConfetti +Bigblumix-BaseFinetune at 1.0. (The order matters!)

Artist tokens work best before the score tags, but you can safely use them at the end

You can find the list of artists here(updating), and just the list of triggers here, style mixing works just have to play with weights, also you can put them in negatives to further guide how you want it to look. Emphasis mode No Norm is suggested for SDXL models.

Huge thanks to @Anzhc with his help in training hyperparameters, to @FallenIncursio , @mfcg, @mnemosynekotorin , @richyrich515 , @justTNP, and @Kojimbomber they were kind enough to provide much needed data for this!

If you want bigger girls go and use mnemosynekotorin's Venus Body lora

Positive prompt : score_9, score_7_up
Helper prompts : score_8_up for photorealistic shading (Zumi, eu03() adviced but not needed, source_anime to make it closer to anime

Negative prompt : monochrome, 3d (when not desired), username, watermark, signature

Personal schizo Negative prompt: worst quality, low quality, simple background, white background, sketch, normal quality, jpeg artifacts, depth of field, blurry, bloom, messy drawing, amateur drawing, fewer digits, extra digits, source_pony, source_furry, source_cartoon, signature, watermark, username, 3d, monochrome

Sampler : Euler A AYS, Any Align Your Steps even GITS.

CFG : 4 to 7 (can be pushed up to 9 and it will work on a lot of prompts

If you want to support me, consider liking the model, and/or tipping buzz/donating to my Ko-fi, as this was finetuned on 8xA100 (Rought stimate is 270 USD for this besides tagging,and testing locally before training)


Relevant training information:

1. Trained with at least 70k images.

2. Total of 3.800.000 real steps after batching across 50 epochs.

3. Effective batch size was 45 x 8 GPUs x 2 Gradient Accumulation

4. Learning rate was 2.5e-6, TE1 was 1e-7, TE2 was 8e-8