It’s been a long time since I came across the AI toolkit by Ostris and a lot of new models have been released. So I decided to jump back into it to train my Desi Babes (for Flux) on Z-Image Turbo using the toolkit.
To run this go to my default setup runpod and subscribe to one RTX5090 Which only costs $0.89 per hour, which is extremely cheap, so I don’t have to worry about stressing my RTX4080 locally and running out of memory.
Ostris has released its entire app as a Docker image that is available as a template on Runpod. Just search under Templates and Keys in Ostris, you will find it.
Once you have your pod running, launch the toolkit. Navigate to Datasets and create a new dataset, upload your images and caption them. If you need help with captioning, check out some of my videos to help with this. Search my ComfyUI Workflow repository for the “Caption” workflow.
Create a new task and set it up with the following settings, many of them default. If you are using more than 24GB of VRAM you can keep the same settings.
The main changes I made: Turn off – Less VRAM, set Quantization Transformer to None, set Max Save Steps to 8 (I like to have more files for testing), Turn on – Training > Cached Text Embedding. The remaining settings are default.
Now you’re ready to get started and get it going. With the RTX5090, I was able to complete the training in under 1 hour and 10 minutes. Checkpoints were available and I could download different versions, as a set it was set to save 8 different versions, I had those and the final version at 3000 steps.
Once you download these you can place them in the ModelsLoras folder in ComfyUI and then you can start testing different versions.
Here are the results of the training. Very happy with the results. I used a LoRA strength of 0.8-0.95.
If you would like to support our site please consider buying us ko-fi, take a product Or subscribe. Need a fast GPU, get access to the fastest GPU for less than $1 per hour RunPod.io
