ZELDA 64 Style - Z-Image/Klein 9B
This LoRA unleashes Nintendo 64 graphics, specifically textures, colors, and lighting from the Zelda 64 titles Ocarina of time and Majora's Mask. Works great for both Z-Image and Z-image-Turbo. I used a dataset of 78 screenshots I took while playing Ship of Harkinian (OOT and MM versions) and captioned them using natural language generated with Google Gemini. Trained with OneTrainer.
The focus here is on Texture Variety. You'll have a hard time trying to create specific characters from the games. It can sometimes be achieved but will typically need detailed descriptions of the character you want to depict.
Suggestions
Begin prompts with 'Zelda64 render' as your stylistic anchor for the most consistent look. For the standard Z-Image version, avoid using the low poly tag because the 64-bit geometry is already part of the base weights and adding it can negatively impact your composition. However, for Z-Image-Turbo, the high speed can sometimes cause the style to look too smooth or modern. In those cases, you should include the low poly tag and increase the LoRA strength to 1.1 or 1.2.
Generations may sometimes try to produce UI elements. Try using tags such as HUD, Health Bar, text, etc. in the Negative Prompt to help prevent this.
Using 'Anime Eyes' in prompts helps to trigger a flat, vibrant look to characters that stays true to the style if that is important for your goal.
LoRA Stacking is highly flexible. It plays well with other character, pose, or outfit LoRA's without shattering the N64 logic.
Responds well to both natural language and tags.
Z-Image Settings
Strength: 0.8-1.0
Sampler: res_multistep or Euler
Scheduler: Simple
Steps: 25–35
CFG: 4.0 I found this to work best.
ModelSamplingAuraFlow: 3.0
Z-Image-Turbo Settings
Strength: 1.0-1.2
Sampler: Euler
Scheduler: Simple
Steps: 8–12
CFG: 1.0
ModelSamplingAuraFlow: 6.0
Klein 9B Settings
Strength: 1.0-1.2
Sampler: LCM/Euler
Scheduler: sgm_uniform/Simple
Steps: 5–9
CFG: 1.0
Resolution: Tested mainly at 832x1216 and 1024x1536. It's very flexible across different aspect ratios.
Feel free to experiment with these settings; they were what I settled on during my initial testing.