Tracer (Overwatch 2) FLUX LoRA
Early access on Patreon (support): https://www.patreon.com/mm744
Crafted from original, high-quality training data at a resolution of 1024x1024. This LoRA model is designed to help you create stunning images for your gallery.
If you enjoy this LoRA, consider supporting it. Your contributions will help in generating the best visuals for the model's showcase. Thank you!
Model Pros and Cons
V1 Model
Pros: Good for pose generation; requires minimal prompting for favorable results; no trigger words necessary*.
Cons: Facial outputs are similar to produced by the general FLUX.Dev model.
V2 Model
Pros: Quality facial generation increased; improved positional character accuracy.
Cons: Requires more precise prompts to prevent random artifacts.
V3 Model
Pros: More precise train settings, the model type is closer to V1 but faces generation much better
Cons: Requires more precise prompts to avoid random artifacts.
Description
Changelog:
Enhanced Face Generation Quality
Expanded Model Size (Incorporating more data)
Tips:
Recommendation: Crafting more precise prompts is advised for optimal results.
Example:
Good: "Tracer stands on the platform, looking directly at the camera, with one hand on her body."
Bad: "Tracer standing on the platform."
Training Configuration:
Images: 41 (1024x1024 each)
Linear: 32
Linear Alpha: 32
Batch Size: 1
Steps: 2250
Noise Scheduler: "flowmatch"
Optimizer: "adamw8bit"
Learning Rate: 1e-4 (0.0001)
Save Dtype: float16