MJ7 Update:
The MJ7 dataset was recreated using MidJourney V7.
It has a better versatility, but it is not as strong as the previous 2 versions.
You can start from higher LoRA strength/weight when using it.
The Lora is trained on MidJourney Images that use my "Personalised Style" + Impressionism
V2 is highly flexible, but the effects are not as prominent as V1.
Works without the trigger words ArsMJStyle, Impressionism
Will start adding visible effects at 0.4+
Works great in the 0.6 - 1.5 range, depending on your preference
It depends on the model, but you can go pretty high with this Lora's weight.
Quality tags and negatives reduce the effect.
Description
The model works great in the 0.4 - 1.5 strength range.
I personally prefer using 1.5 strength on the Flux Dev GGUF Q4
FAQ
Comments (4)
God Almighty thinks this still needs a bit of work. I am not sure Money isn't rolling in his grave right now. But it's a good idea!
The Flux Model does not recognizes the Artists. This is the point of that showcase. Maybe I should have stated this more clearly. I never use actual art pieces in my training data - only synthetic Data.
Great work on this LoRA! Agree with OP that best artistic outputs are around 1-1.5 LoRA weights and some playing around with the Flux Guidance between 2-3. Thanks again for this!
Thanks 😁 And thank you for the comment! I also like the 1-1.5 range, as I mentioned in the version description 1.5 seems like the perfect spot for me in GGUF Q4 Dev. I don't recommend CFG ranges as it is too model Dependant and I expect alot of fine tunes and LoRA merged models to appear in the next weeks 😁
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.