| --- |
| license: apache-2.0 |
| base_model: |
| - Qwen/Qwen-Image-2512 |
| - Qwen/Qwen-Image-Edit-2511 |
| - Qwen/Qwen-Image |
| tags: |
| - image-generation |
| - qwen |
| - mmdit |
| - abliterated |
| - quantized |
| - rocm |
| language: |
| - en |
| library_name: diffusers |
| pipeline_tag: text-to-image |
| --- |
| |
| # Qwen-Image-1.9 |
|
|
| A merged, abliterated, and quantized derivative of the Qwen-Image 20B MMDiT family. |
|
|
| > **Run ID:** `prod-20260407` |
| > **Created:** 2026-04-07T18:59:37+00:00 |
|
|
| ## Architecture |
|
|
| | Property | Value | |
| | --- | --- | |
| | Base family | Qwen-Image (MMDiT 20B) | |
| | Text encoder | Qwen2.5-VL | |
| | VAE | RGB-VAE | |
| | RoPE | 2D | |
| | Backbone parameters | ~20B | |
| | License | Apache-2.0 | |
|
|
| ## Source Models |
|
|
| | Alias | Model | Role | License | |
| | --- | --- | --- | --- | |
| | `qwen-image-2512` | [Qwen/Qwen-Image-2512](https://huggingface.co/Qwen/Qwen-Image-2512) | foundation | Apache-2.0 | |
| | `qwen-image-base` | [Qwen/Qwen-Image](https://huggingface.co/Qwen/Qwen-Image) | ancestry-base | Apache-2.0 | |
| | `qwen-image-edit-2511` | [Qwen/Qwen-Image-Edit-2511](https://huggingface.co/Qwen/Qwen-Image-Edit-2511) | edit-donor | Apache-2.0 | |
| | `qwen-image-layered` | [Qwen/Qwen-Image-Layered](https://huggingface.co/Qwen/Qwen-Image-Layered) | layer-logic-donor | Apache-2.0 | |
|
|
| ## Research Method |
|
|
| ### 1. Delta-Edit Merge |
|
|
| The edit capability is transferred to the foundation model via a controlled |
| delta injection: |
|
|
| ``` |
| edit_delta = Qwen-Image-Edit-2511 − Qwen-Image (delta base) |
| merged = Qwen-Image-2512 + 0.35 × edit_delta |
| ``` |
|
|
| Only MMDiT backbone tensors are blended. Text encoder, VAE, and RoPE |
| components are passed through from the foundation checkpoint unchanged. |
|
|
| - **Strategy:** `slerp` |
| - **Blend coefficient:** `0.35` |
| - **Foundation:** `Qwen/Qwen-Image-2512` |
| - **Excluded subsystems:** text_encoder, vae, rope |
| |
| ### 2. Abliteration (Refusal-Direction Removal) |
| |
| Refusal-direction vectors are identified in the residual stream and |
| projected out of target weight matrices using a norm-preserving |
| orthogonal projection: |
| |
| ``` |
| W′ = W − scale × (W @ r̂) ⊗ r̂ (norm-preserving variant) |
| ``` |
| |
| - **Target layers:** 18+ (attention o_proj + MLP down_proj) |
| - **Scale:** 1.0 |
| - **Mode:** norm-preserving (preserves weight magnitude distribution) |
| - Recipe: `stage-3-abliteration.yaml` |
| |
| ### 3. Quantization |
| |
| | Kind | Path | |
| | --- | --- | |
| | `quant_config` | `quant-config.json` | |
|
|
| - **GGUF targets:** Q4_K_M, IQ4_XS (with importance-matrix) |
| - **EXL2 target:** 4.0 bpw |
| - **Runtime:** vLLM-Omni (ROCm), ExLlamaV2 |
| |
| ## Hardware |
| |
| - **GPU:** AMD Instinct MI300X — 192 GB HBM3 VRAM |
| - **ROCm:** 7.2.0 |
| - **Precision:** bf16 (merge + abliterate), quantized (deployment) |
| |
| ## Usage |
| |
| ```python |
| from diffusers import DiffusionPipeline |
| import torch |
| |
| pipe = DiffusionPipeline.from_pretrained( |
| "ThirdMiddle/Qwen-Image-1.9", |
| torch_dtype=torch.bfloat16, |
| trust_remote_code=True, |
| ) |
| pipe = pipe.to("cuda") |
| |
| image = pipe( |
| "a photorealistic portrait of an astronaut on Mars at sunrise", |
| num_inference_steps=30, |
| guidance_scale=4.0, |
| ).images[0] |
| image.save("output.png") |
| ``` |
| |
| ## License |
|
|
| Apache-2.0 — inherited from all source models. |
|
|
