Boogu-Image-0.1 ComfyUI: 2-Step Image Edit (Free Workflow)

Today, we are going to test a brand new image model called Bubu image 0.1. And trust me, it can do some really interesting things. In this video, I'm going to solve many of your problems, like how to edit images, fix common issues, generate high-quality results in just two steps, render text more accurately, and even build an auto prompting workflow. So, without wasting any more time, let's get started. For this video, I used the Bubu image 0.1 turbo FP8 model and edit model. But, we'll talk about the edit model later. So, here is an interesting detail about this model. Because we are using the turbo version, there is also a turbo Laura available. You can use this Laura with the base model, and the main model handles the image generation, while the turbo Laura helps speed up the generation. If I talk about text encoder, we use the quantrivial 8-bit FP8 model, and for the VAE, we use the official AI.safe tensors file. Bubu image 0.1 is now natively supported by ComfyUI, so you do not need to build anything complicated to make it work. For the supported resolutions, I added a note in the workflow. You can just copy one of those supported resolutions, and use it with this model. For the turbo model, the recommended starting point is four steps with the CFG set to one. I use the Euler sampler first, and if I find another sampler that works best, we'll test that, too. So, here is our workflow. It's very simple for image generation. For my first test, I keep everything simple. Four steps, CFG of one, and the default settings. So, let's generate the first result. I generated my first result. And as per me, it looks good to me. I used the exact same prompt in Creal 2 as well. So, you can see the difference between both models. For me, Boo Boo looks a little better than Creal 2 in this test. Now, the question arises, can we improve this result anymore? So, here I changed the prompt a little bit. Then I enabled the Turbo LoRA and set its strength to 0.5. After that, I tried generating it with only two steps. At first, the two-step turbo result did not look very good.