qwen3_4b_Rayzist_v1.0 (FP8)
Qwen3-4B encoder merged and trained for use with Rayzist — the Z-Image Turbo finetune from Ray's Models Zoo. Quantized to FP8 scaled.
Base Model
Built on Qwen/Qwen3-4B — Alibaba's 4B causal LM with thinking/non-thinking mode switching, 32K native context, and strong multilingual capabilities (Apache 2.0).
Rayzist
Rayzist is the Z-Image Turbo model family from MutantSparrow/Ray, optimized for fast inference, text rendering, and multi-style generation. Built on a hand-curated dataset of original street photography with a natural, analog-leaning aesthetic.
This encoder is trained specifically to pair with Rayzist — visit the model zoo for the full pipeline and recommended settings.
Usage
Use alongside the Rayzist Z-Image Turbo pipeline. See MutantSparrow/Ray for full setup instructions.
License
Apache 2.0 — inherits from Qwen3-4B base.