UrangDiffusion V3.1 - V3.1

UrangDiffusion XL v3.1 is fine-tuned from Animagine XL 4.0 Base (not Zero). This 4.0 Base model serves as the base model pre-trained for the final release of Animagine XL 4.0 (not the Opt version).

I have received permission from the team to fine-tune the base model using my own method and release it under the UrangDiffusion series.

Base model: Animagine XL 4.0 Base
Fine-tuning details:

• Dataset size: ~1,600 images

• GPU: 1× A100 80GB

• Optimizer: AdaFactor

• UNet learning rate: 1.25e-6

• Text encoder learning rate: N/A (disabled)

• Batch size: 48

• Gradient accumulation: 1

• Warmup steps: 5%

• Minimum SNR: 5

• Epochs: 15

Due to some quirks of the model, please keep the following in mind:

• v3.0 may perform better with anatomy

• v3.1 may perform better with more fluid poses

If you encounter anatomical issues at 28 steps, try lowering to 27 or increasing to 29. If it improves but isn’t perfect, continue adjusting slightly up or down. If the result worsens, the previous step count was likely the optimal one.

UPDATE [19/04/2025]

• Some generations are stable at step 30++ with Euler a. You might wanna crank up the steps a bit.

• Some generations are also better with realistic, 3d included in the negative. Try that too.