V1.0 Turbo:

I have added the model 1.0 turbo, is it meant for low step counts while minimizing issues. The main base for the was the flux-turbo-alpha lora I merged in. I then proceeded to finetune that and optimize the weights to create a model that preforms incredible at low step counts. It can often be a 3x increase in speed for a somewhat minor drop in quality. I believe for those who are gpu poor this is the best high speed realistic model you can get and I am extremely happy with its results

This is a model I built that works to finetune the original flux krea model. In every prompt I tested my model far outperforms. The model has a noticable jump in detail, lighting, and prompt adherence compared to the base flux krea. I have gotten some extremely high quality shots with this model that I have not seen anywhere else

I recommend 28 steps to avoid grainyness caused by the model with poor prompts. 20 steps also works most of the time but a simple one or two word prompt might look ugly. I have noticed an issue where the skin detailers get overtuned and sometimes like to create skin disorders, a good prompt fixes this though. I will try to train this out in the future. This is my first model so it may not be perfect. I am going for realism above all else, this means the model is less artsy than the normal flux krea but it does feel more realistic.


Below is the same prompt and seed as the title with the base flux.1 dev krea (FP8):