John2386/fullgreed-edit

Fullgreed Edit

Fullgreed Edit is a Z-Image-Edit-style Diffusers repository. The transformer weights were created by transferring the Fullgreed delta onto the Z-Image-Edit donor transformer architecture:

Z-Image-Edit + (John2386/fullgreed - Z-Image base)

What is included

• Fullgreed edit transformer: transformer/diffusion_pytorch_model.safetensors
• Z-Image-Edit transformer config with in_channels: 32
• Qwen3-4B text encoder files from the edit donor
• tokenizer files from the edit donor
• Flux/Z-Image VAE files from the edit donor
• FlowMatch scheduler config from the edit donor
• Diffusers model_index.json

Important note

This is not a paper-scale retraining of Z-Image-Edit. It is a Fullgreed delta transplant into the Z-Image-Edit donor architecture, with the missing non-transformer runtime components mirrored from the donor repo.

A true paper-style model would require continued edit training on paired source/instruction/target images. This repo is intended to preserve the donor edit structure while carrying the Fullgreed visual style.

Smoke testing

Use a Z-Image-Edit-compatible runner. The current public Diffusers ZImagePipeline support may lag behind the paper-style source-image editing API, so if a loader exposes only text-to-image generation, use the donor/edit runner that supports reference-image latents.