I've been experimenting with this model, but I'm taking a break now to focus on other stuff.I used the Grok 4.1 inference model to tag the images, gave them a quick check and tweak, and then started the training.I picked some images to train the "anal x-ray" concept. Turns out, the image flipping feature was really messing things up—it was causing broken limbs and making the prompt descriptions not match the pictures. Once I turned that off, the model started working and converged much faster.It made me wonder if image augmentation is even worth doing anymore. I think having better, more structured data and accurate captions is way more important.Since I fixed the bug late, I ran the training for a few extra steps to compensate.I started with 5,000 steps, a 2e-4 linear learning rate (LR), and a Batch Size of 4. Later, I switched the LR to a cosine decay (from 2e-4 down to 4e-5) and doubled the batch size to 8. I stopped at 12,000 steps after watching the samples come in.I ran the whole thing on <a target="_blank" rel="ugc" href="http://vast.ai">vast.ai</a> using the ai-toolkit and an RTX Pro 6000 Blackwell (mostly just to try out a bigger batch size).You could try some related concepts, like...?It is recommended to reduce the weights to avoid overfitting; the weights should be between 0.8 and 0.9, and not exceed 1.

anal x-ray (NetaYume Lumina 3.5)

This is an experimental version, so I'm still working through some issues.Overall, it feels limited in how deeply it can break down complex concepts. This might be due to the LLM’s inherent capabilities or a problem with the text embeddings the model uses.But its ability to align with longer prompts is clearly better, allowing it to surpass the limits of SDXL.However, the arrival of Z-image changes everything. Maybe Lumina isn’t the SDXL successor we were actually waiting for.

Files with SHA256: a48da2fab267b3b0bb96697d004cc9caa3ef233f628ed0d0e332ee5bdf5cb3fc