If you have found a legitimate .safetensors file labeled "anya_oxi_patched_v4.safetensors," follow this installation guide for Automatic1111 or ComfyUI.
Step 1: Backup Your Original Model
Before replacing files, move your old anyaOxi.ckpt to a backup folder. The patched version uses a different hash; do not just rename the old file.
Step 2: Download and Place the File
Step 3: Select the Correct VAE Unlike the original, the patched model requires an external VAE. anya oxi model patched
Step 4: Recommended Settings Based on community testing (Civitai, November 2024), use these parameters for the best results:
The most critical patch involves the Variational Autoencoder (VAE). The original Anya Oxi had a "baked-in" VAE that was incompatible with standard EMA (Exponential Moving Average) weights. This caused magenta flashes in high-contrast images. The patched version decouples the problematic VAE, making it fully compatible with standard 840000 VAE files.
To understand the patched version, you must first understand the original. The Anya Oxi Model (often improperly trademarked as "Anya OXI" or "Anya OXP") was a custom Stable Diffusion checkpoint. It was celebrated for several specific traits: If you have found a legitimate
Despite its popularity, users quickly discovered a fatal flaw. The original 2.0 and 3.0 variants suffered from what the community called the "glassy artifact" or "latent bleeding"—specifically, a tendency to bake unwanted noise into the background (resembling oxidized rust or static) when using high-resolution fix or CFG scales above 7.
The original model over-indexed on its "oxidized" training data. When generating simple prompts like "a girl sitting in a room," the background would automatically generate rust spots or water stains. The patched model keeps the aesthetic color palette but removes the environmental decay artifacts.
| ID | Type | Severity | Impact |
|----|------|----------|--------|
| CVE-2026-0142 | Prompt injection via special token <|oxi_end|> | High | Unauthorized read of system prompt & partial chat history |
| Internal-134 | Tokenizer collapse on repeated Unicode combining characters | Medium | OOM on 24GB GPUs with 64k+ context | Step 3: Select the Correct VAE Unlike the
The creator of the original Anya Oxi (who remains pseudonymous as "Oxi_Diffuser") has reportedly abandoned the project following the patch controversy. However, a collective of maintainers (calling themselves "The Anvil Team") has taken over development.
They have announced that the Anya Oxi Model Patched will be the final version of the SD 1.5 lineage. Future development will shift to SDXL and Pony Diffusion V7 architectures. This means the current patched model is the definitive, final version for SD 1.5 workflows.
The Anya Oxi Model Patched (Version 3.2P and 4.0P) is not merely a renaming; it is a fundamental surgical correction of the original model’s latent space.
Here is the technical breakdown of what the patch actually fixes: