Showing you the maximum potential that zit/base can achieve
FP32 model achieves photorealistic images without LoRA or tiled upscaling...
An anonymous developer has shared a series of image samples demonstrating the ZIT/B model's claimed superiority in realism over all competing image generation models, including both open-source and closed-source alternatives. The post emphasizes that these results were achieved using only the base FP32 version of ZIT or ZIB, without any LoRA modifications or tiled upscaling. Instead, the resolution was increased in a single pass to the maximum value the model can withstand before crashing, using a technique called dype. All prompts were written with GPT to ensure optimal output.
The developer states they purchased a Pro6000 at their own expense for research and have not earned any money from the model. They may keep the companion WF variant private, as it represents months of testing effort. The post includes 18 sample images hosted on postimg.cc, showing faces and scenes with high photorealism. The developer concludes by asserting that ZIT/B has no rivals in realism and that future comparisons should reference this demonstration.
- ZIT/B base FP32 model achieves photorealistic results without LoRA or tiled upscaling
- Single-pass resolution upscaling to the model's crash limit (using dype technique)
- Developer funded a Pro6000 themselves and may keep the WF variant private
Why It Matters
ZIT/B could set a new benchmark for image realism, challenging both open-source and proprietary models.