Disclaimer: This model is a little out-of-control. But dammit, I love it to death. It's intentionally kind of unhinged, and loves creating colorful, dreamlike, surreal illustrations. Don't use this model expecting a strong general-purpose experience. I can't make any guarantee that this model will be clean and coherent. This model was a creative exploration with the goal of a certain strong vibe and wacky creativity, not raw power and cleanliness. Leans a good bit towards its own illustrative style, so alternative styles may be trickier.

To get the full power of this model's creativity, I strongly recommend limiting your negative prompting as much as reasonably possible. You can use the bad-artist negative embedding to achieve certain effects, but beyond that embedding, I don't condone a lot else unless you are very sure of what you're doing.

If you're still interested by this model, read on.

SOME USAGE TIPS AND INFO:

• I recommend keeping CFG around 7 and below. 5-7 is always a safe bet.

• All my images were generated with 30 steps DDIM and upscaled via img2img at 0.47 strength (High Res Optimization, for InvokeAI users). You may need more steps if you aren't upscaling through that method.

• Put in OliDisco style at the start of your prompt for interesting, trippy effect. Works best on vague as well as short prompts.

• To kill unprompted extra random colors, put rainbow in your negative prompt, and weight as needed.

• To kill the light glowing effect, put backlighting in your negative prompt, and weight as needed.

• You can also try triggering the small portion of flat illustration LoRA with flat illustration in your prompt, the Vector Graphics Art Style LoRA with flat color in your prompt, and the Dreamy Vector Style LoRA with vectorstyle. Note those LoRAs were mainly used to enforce a style, but you can still get interesting effects with them.

MODEL BACKGROUND:

This model was based off of my 526Mix-Animated mix. While I would have loved to use my trusty ol' 526Mix, the Animated version has better lighting introduced by Nerfgun3's Macaron Mix that I felt would be more in line with the effects I wanted.

To create this model, I merged a bunch of LoRAs on top of 526Mix-Animated. This was done in two parts, with the goal of expanding on 526Mix/526Mix-Animated's natural creativity, while creating some nice cozy and thoughtful vibes.

Full credits to the authors of these LoRAs, and I suggest using them on their own to get their full powers.

Part 1:

• AlchemistW's Enhanced Backlighting at 0.40 multiplier

• IronCat's Anime vibrancy & simplifier at 0.40 multiplier

• TLME's Vector Graphics Art Style (No lineart) at 0.80 multiplier

• olivetty's Soul of Disco Diffusion at 0.60 multiplier

This model made stunningly bright and colorful illustrations, but was often a bit too saturated.

Part 2:

• AlchemistW's Enhanced Backlighting at 0.50 multiplier

• l616631804190's flat illustration at 0.65 multiplier

• olivetty's Soul of Disco Diffusion at 0.60 multiplier

• The_Jingo's Cosmic Rainbow at 0.25 multiplier

• Gertan's Dreamy Vector Style at 0.50 multiplier

Parts 1 and 2 were then mixed at a 50/50 ratio to balance each other's quirks.

Judging by the time I typed 'that dumb LoRA has me wanting to make an unhinged surrealist model...' in the InvokeAI discord server, total experimentation time was about 11 hours, not including the several hours of pre-exploration before going through on the idea for this model.