ARTSY VIBE
Important thinks to know about Artsy Vibe:
- It doesn't prompt the same way as base flux.1-Dev. It wants you to be wordier and more descriptive, but yields more detailed results.
- My examples are cherrypicked. I'm not going to pretend they aren't. Vibe has a tendency to screw up anatomy. But this is a minor cost for the advantage of the creative, stylistic and detail changes. Don't be afraid to hit 'run' a few times before you find that gem.
- While Artsy vibe is capable of some pretty decent realism, it is intended as an artistic model that often, like it's predecessor "Artsy Dream", leans into the uncanny nature of AI image generation rather than avoiding it.
VERSION 2 - (only fp8 & fp16 this time)
The long overdue second installment of Artsy Vibe is not as polished as I expected it would be. Some styles may require more iteration. Honestly, I'd say it leans more experimental and less predictable. However, with a greatly improved variety of 2D styles, more unique poses, and vastly improved background detail, I think people will find it to be a great step forward.
You'll find that some styles are almost exactly the same as version 1, while others are substantially changed. Hopefully version 3 will be much less of a wait than version 2 was.
Thanks for the support, and enjoy!VERSION 1
The next step in my series of ARTSY models. As a spiritual successor to Artsy Dream, this model will also lean into uncanny world of AI art rather than try to resist the unavoidable. Artsy Dream achieves this with a wonderful set of personal aesthetic preferences, most of which are as refined as they can be as a Flux.1-Dev based model. However, this new model marks a substantial difference in aesthetic direction, and a new project of exploration and refinement begins now. In this first version, I define the basic styles of the project. Though, admittedly, more is left to be desired. Stay tuned for future updates as this project develops!
This model performs spectacularly with realism, anime, and various illustrative styles. I'm delighted to share that it produces a large variety of text styles with great stylistic prompt adherence. Expect over-the-top compositions and vivid scenery.
Enjoy!VERSIONS AVAILABLE:
FP16 (.safetensors)
FP8 (.safetensors)
SOME GGUF QUANTS MAY NOT BE FULLY FUNCTIONING IN FORGE UI DUE TO COMPATIBILITY ISSUES CURRENTLY BEING ADDRESSED BY ILLYASVIEL.
These models use the same clip_l, t5xxl (FP16 or FP8), and ae.safetensors as Base Flux.1-Dev & Flux.1 Schnell.
(Images attached to each version have been generated with their respective model sizes/quants for accuracy and honesty. Though you'll be surprised at the level of detail achievable with this Q4_K_S.)
REGARDING NSFW: I do not add any NSFW data to my models. It can produce NSFW material in about the same capacity as Flux Dev, and I will not be adding any NSFW capabilities to any future versions.
Description
FAQ
Comments (12)
oh damn i missed the release of v2 of this, thank you, the fp16 model works great on my old 3090 too,
Oh come on, 3090 is still a powerhouse!
I put this thing together on my 3080ti!
though LoRAs merged in were trained either here on civitai, and some on replicate.
After doing some seed comparisons between v1 and v2 models, I feel like v2 is more prone to creating oddities. For example, v1 had a water bottle behind a plant, in v2, the bottle and plant were weirdly merged together. It's still a great quality model, and maybe if you want some abstract type artworks, it would be better for it, but for me the lack of stability made it feel down a downgrade. Thanks regardless for your work, I really love your models!
This is totally true, and I made mention of it in the description! Thanks for the feedback.
For certain styles it introduces some wild artifacting, requiring some more iteration. It will definitely be a downgrade for those who aren't taking advantage of the improved areas.
That said. v3 is probably not too long down the road...
I get this error using forge:
You do not have VAE state dict!
And ive tried too load a clip and t5, still same error
What version of the model were you using? Forge has compatibility issues with some GGUF files. I recommend the fp8.
Anything new in the works? This is my favorite Flux model.
I seriously appreciate your support, unfortunately I'm not in the business of sharing custom models anymore.
@jurdn shame, tip of the iceberg and all...
@AbsoluteReality Not for this community. Looking more like the end of the road.
Can you use this with Forge? and if so which VAE/encoder do you use?
how does one get this suitable for nunchaku, i keep seeing gems of generations and i want to try it with my current workflow D;