<h3 id="this-is-a-short-guide-how-you-can-train-the-brand-new-krea-2-raw-loras-in-high-quality-with-low-budget-hardware-by-using-ostris-ai-toolkit.">This is a short guide how you can train the brand new Krea-2 raw Loras in high quality with low budget hardware by using Ostris AI Toolkit.</h3>The created Krea-2 raw Loras you can use with Krea-2 raw models as well with the much faster Krea-2 Turbo model.By using the Turbo model with your trained raw Lora you can generate images very fast and in good quality with 12 GB VRAM too. This works perfectly with comfyui for example.I struggled for several days with Musubi Trainer and other tools to get it running on my low budget system, but it all did not work well.Finally I found the the main parts for a very easy to use solution with Ostris AI Toolkit here: <a target="_blank" rel="ugc" href="https://www.reddit.com/r/StableDiffusion/comments/1uh95sz/training_krea_2_lora_on_rtx3060_12gb_a_slow/?utm_source=chatgpt.com">reddit </a>(all credits go to the author).Because this works very well after some testing and it produces high quality results I would like to publish it here with all setting parameters.Ok - lets go:<h3 id="what-do-you-need:">What do you need:</h3><ul><li>RTX 3060 GPU with 12 GB VRAM,</li><li>32 GB RAM,</li><li>probably a swap-file on a fast SSD: for myself I use a very large swap-file size of 64 - 128 GB (for video generation), but I`m pretty shure it will work with a much smaller size, like 32 GB for example.</li><li>Ostris AI Toolkit (<a target="_blank" rel="ugc" href="https://github.com/ostris/ai-toolkit/">github-page</a>). On Windows just use Travis1 <a target="_blank" rel="ugc" href="https://github.com/Tavris1/AI-Toolkit-Easy-Install">Easy-Install.</a></li><li>my settings (use the atteched config.yaml file or the screenshots).</li><li>around 20 - 30 high quality images for a character Lora for example.</li></ul><h3 id="how-long-will-the-training-run:">How long will the training run:</h3>A complete character training for a high quality Lora with 750 training steps will take around 8 - 9 hours. So this works well as an overnight job.<h3 id="what-are-the-limitations-with-12-gb-vram:">What are the limitations with 12 GB VRAM:</h3><ul><li>the training resolution is limited to 512+768. 1024 work too, but training will need several days. So 512+768 works well and delivers very good results with Krea-2 Turbo image generation.</li><li>rank 24 insted of 32 (I don't think it makes much of a difference).</li><li>Learning Rate = 0,0002 insted of 0,0001 to reduce the steps (training time to 50%) - see below.</li></ul><h3 id="preperation:">Preperation:</h3>Training Images (aspect ratio, resizing):Use congruent, sharp high quaility images (at least 1024 x 1024) of any aspect ratio. Do not downscale your images. AI Toolkit creates automatically buckets.Captions:Character training works well without any captions. But if you like, you can use captions too.<h3 id="how-to-start:">How to start:</h3>Open AI Tollkit and create a new Dataset with your images (and optional with your captions). Create a new Job and use my settings from config.yaml or the screenshot.<h3 id="for-a-dataset-without-captions-dont-change-anything-exept:">For a Dataset without captions don`t change anything exept:</h3><ul><li>set your Trigger Word,</li><li>select your Target Dataset,</li></ul><h3 id="for-a-dataset-with-captions-dont-change-anything-exept:">For a Dataset with captions don`t change anything exept:</h3><ul><li>delete the Trigger Word and leave it blank,</li><li>select Cache Text Embeddings (Unload TE gets unselected),</li><li>select your Target Dataset.</li></ul>Some hints:<ul><li>AI Toolkit automatically downloads the correct Krea-2 raw model files after starting the job (make sure you have sufficient free disk space).</li><li>AI Toolkit do Cache Latents and Unload Text Encoder automatically to save VRAM for training.</li><li>We use 750 steps and Gradient Accumulation = 2 - this means we train with 1500 "virtual" steps.</li><li>With Leraning Rate = 0,0002 we get well trained character Loras mostly between 400 and 750 steps. Because we save every 100 steps you finally can choose the best output.</li><li>Do not activate Sampling. This leads to OOM errors.</li><li>The first training steps take very long (around &gt; 300 seconds). Don`t worry. It speeds up with every step and the complete training will take around 8 to 9 hours.</li></ul>

Files with SHA256: a5870f20e110bc6b08a5a3dd3928af1d000fb77b74e7103bd5966c55dda59086