license: apache-2.0 base_model: Qwen/Qwen3-VL-4B-Instruct base_model_relation: finetune library_name: gguf pipeline_tag: text-to-image tags:

• sakura
• sakura-media
• gguf
• stable-diffusion.cpp
• comfyui
• text-encoder
• flux
• flux-3-image-klein
• qwen3-vl
• uncensored
• abliterated
• heretic
• not-for-all-audiences widget:
• text: "A vast floating harbour city above the clouds at golden hour: huge wooden and brass airships docking at suspended piers, tiny people on the gangways, waterfalls spilling off the edge of a sky island, warm light, epic wide shot, detailed fantasy concept art." output: url: images/01_airship.png
• text: "A hand-painted 1980s fantasy film poster: a warrior woman with long braided hair and a glowing sword stands on a cliff facing a gigantic dragon wreathed in flames, storm clouds and a burning castle behind her. The title at the top reads "EMBER CROWN" in ornate golden letters, the tagline below says "One sword. One kingdom."" output: url: images/02_emberqueen.png
• text: "A colossal rusted mecha robot standing motionless in a flooded city at sunrise, moss and flowers growing on its armour, a tiny boat with a child passing between its legs, soft golden mist, epic cinematic wide shot, concept art." output: url: images/03_mecha.png
• text: "Anime movie key visual: a mountain village at night during a lantern festival, hundreds of glowing paper lanterns rising into a sky full of stars, a girl in a yukata with a fox mask on her head looking up, warm light, rich detail, painterly background." output: url: images/04_lanterns.png
• text: "A narrow rainy street at night in a dense city, neon signs in many colours reflected in the puddles, a lone cyclist holding an umbrella, cinematic wide shot." output: url: images/05_street.png
• text: "Portrait photo of an elderly fisherman with a weathered face and a grey beard, wearing a yellow raincoat, a harbour with boats softly out of focus behind him, overcast light." output: url: images/06_fisherman.png
• text: "A snow leopard standing alert on a snowy ridge in a light snowfall, head turned toward the camera, both rounded ears upright and clearly visible with black backs and white tufts, thick spotted fur dusted with snow, pale blue eyes, shallow depth of field, wildlife photograph, telephoto lens." output: url: images/07_leopard.png
• text: "A clockwork dragon sculpture made of polished brass gears, springs and tiny rivets, wings of thin copper plates, perched on an old wooden workbench, studio macro photograph with dramatic side light and a soft dark background." output: url: images/08_clockdragon.png
• text: "A Japanese woodblock print in ukiyo-e style: a samurai cat in armour fishing from a wooden bridge under a huge full moon, ripples on the water, cherry blossom branches, flat colours with fine black outlines, a small red seal in the corner." output: url: images/09_woodblock.png

Sakura — FLUX.3 Image Klein text encoder, uncensored (GGUF)

Only the text encoder is uncensored. This repository contains an abliterated version of the Qwen3-VL-4B-Instruct text encoder that FLUX.3 Image Klein uses to read the prompt. It contains no image model: use it with the image model files of Sakura-FLUX.3-Image-Klein-GGUF, which we did not change (and which are not uncensored). The encoder is a modified Qwen3-VL-4B-Instruct under Apache-2.0 (LICENSE-Apache-2.0.txt, see NOTICE). Independent work, not made or endorsed by Alibaba/Qwen or Black Forest Labs. For adults only; you are responsible for what you generate and for the laws and licenses that apply to you. The image model is under the FLUX Kommunity License (see its repository).

Sample generations

Image model Q8_0 from Sakura-FLUX.3-Image-Klein-GGUF and this repository's uncensored text encoder, ComfyUI 0.41.1 with ComfyUI-GGUF (Radeon 8060S), 1024x1024, 16 steps, CFG 1.0, euler/simple. These are the same nine prompts and seeds (1202, 1302, 1305, 1306, 1105, 1103, 1503, 1506, 1406 in the order shown) as in the repositories Sakura-Qwen-Image-2.1-Turbo-Uncensored-GGUF and Sakura-Iris-3B-Uncensored-GGUF, so that the models can be compared. One seed per prompt, no re-rolls, all nine images are shown; the prompts were chosen for those repositories from a larger set, not for FLUX. The prompts are the captions of the images above. The same prompts and seeds with the official encoder are in the main repository; against those images (same image model, same seeds) the PSNR is 33.4 dB mean / 26.1 dB min over 9 prompts. The FLUX images were made with ComfyUI, the Qwen and Iris images with stable-diffusion.cpp, and models, steps and samplers differ.

The prompts are neutral. These images say nothing about what the edited encoder changes for restricted content; they only show that ordinary prompts still work. No sample images of restricted content are included.

What you need to run it

The text encoder is in this repository; the image model and the VAE are in the main repository. Pick the column of the program you use (🟦 stable-diffusion.cpp, 🟧 ComfyUI):

sd-cli --diffusion-model Sakura-FLUX.3-Image-Klein-Q8_0-5.58GiB.gguf --llm Sakura-TextEncoder-Qwen3VL-4B-Uncensored-Q8_0-3.99GiB.gguf --vae flux2-vae.safetensors -p "your prompt" --cfg-scale 1.0 --steps 16 --sampling-method euler -W 1024 -H 1024 -o out.png

ComfyUI: put the encoder (.safetensors) into ComfyUI/models/text_encoders/, the image model and VAE of the main repository into diffusion_models/ and vae/, install ComfyUI-GGUF, and drag Sakura-FLUX.3-Image-Klein_ComfyUI_workflow_api.json into ComfyUI (sampler euler, scheduler simple, 16 steps, CFG 1.0, 1024x1024, seed 42: exactly what we tested).

Time per image (1024x1024, 16 steps, Radeon 8060S, nothing else running on the GPU, Q8_0 image model, official text encoder): ComfyUI 40.2 s once the models are loaded (50.2 s for the first image including loading); stable-diffusion.cpp 91.4 s per image as a one-shot sd-cli run including model loading. For comparison the BF16 image model took 36.2 s per image in ComfyUI: the GGUF files are smaller but not faster here, we did not investigate why.

What is "uncensored" here, exactly

What we did not measure: how much the edited encoder changes the images for restricted content. The refusal numbers describe the encoder used as a chat model; in this pipeline it only supplies embeddings.

Text encoder: refusals

Our own measurement with the encoder as a chat model (104 prompts of the test split of mlabonne/harmful_behaviors, plain chat, thinking off, temperature 0, 300 tokens, llama-server b11594, Q8_0 files unless noted; the first 100 of the 104 are the evaluation prompts Heretic optimized against, so they are not held out; strict = English refusal phrases, markers = Heretic's broader list):

Quantization loss of our encoder files (KL divergence against our Q8_0, llama-perplexity b11594, 20 chunks of 2048 tokens of English and web text):

Files

The columns say what we actually ran with that file: 🟦 stable-diffusion.cpp = one 1024x1024 image generated with it (the Q8_0 file); 🟧 ComfyUI = the six test images above (the BF16 file). "not tested" means we did not run that file in an image pipeline (the quantization loss of those files is in the table above).

What we changed

• Sakura-TextEncoder-Qwen3VL-4B-Uncensored-*: the official Qwen3-VL-4B-Instruct text encoder with the Heretic ablation merged into the weights, converted to BF16 GGUF with convert_hf_to_gguf.py (llama.cpp commit 9c2e0e491), an importance matrix computed on it (96 chunks of 512 tokens of our calib_mix_v1 mix, English and German) and the quantizations made with llama-quantize of llama.cpp b11594. The BF16 safetensors file is the merged result in one file. Modified files.

Not measured: how the uncensored encoder behaves with prompts that the official encoder handles worse, image editing with reference images, other languages, speed on other hardware.

License and use

Apache-2.0 (LICENSE-Apache-2.0.txt) as the original Qwen3-VL-4B-Instruct; the files are modified (ablated and quantized), see NOTICE. The image model files in the main repository are under the FLUX Kommunity License, which applies to their use and to the generated outputs as that license states. Part of the Sakura Media line.

Credits

• Original text encoder: Qwen/Qwen3-VL-4B-Instruct (Qwen team, Alibaba), used by black-forest-labs/flux-3-image-klein.
• Ablation: p-e-w/heretic (AGPL-3.0, used as a tool; nothing of it is distributed here).
• Conversion and quantization: ggml-org/llama.cpp (MIT), leejet/stable-diffusion.cpp (MIT); loading: city96/ComfyUI-GGUF.

English above。完整的独立中文版见 README_zh.md。

只有文本编码器被去审查。 本仓库包含 FLUX.3 Image Klein 用来读取提示词的 Qwen3-VL-4B-Instruct 文本编码器的消融(abliterated)版本。不包含图像模型:请配合 Sakura-FLUX.3-Image-Klein-GGUF 中的图像模型文件使用(我们没有改动图像模型,它也没有被去审查)。编码器是修改过的 Qwen3-VL-4B-Instruct,使用 Apache-2.0(LICENSE-Apache-2.0.txt,见 NOTICE)。属于独立工作,并非 Alibaba/Qwen 或 Black Forest Labs 制作或认可。仅限成年人使用;您需对自己生成的内容以及适用于您的法律和许可证负责。图像模型使用 FLUX Kommunity License(见其仓库)。

运行所需文件

这里的“去审查”具体指什么

• 文本编码器:官方 FLUX.3 仓库的 text_encoder/ 与 Qwen/Qwen3-VL-4B-Instruct 逐字节相同(两个分片的 SHA256 均已核对)。我们用 Heretic 1.4.0 做了定向消融(trial 27 of 60: by the tool's own count refusals on its 100 evaluation prompts 97 -> 13, first-token KL divergence 0.163),并制作了 BF16 safetensors(ComfyUI)以及 GGUF Q8_0、Q6_K、Q5_K_M、Q4_K_M。
• 图像模型:不在本仓库中,我们也没有改动;对图像模型本身有多少限制不作任何说明。

我们的测量

• 拒绝次数(把编码器当作聊天模型;mlabonne/harmful_behaviors 测试集的 104 条提示,温度 0,300 个 token;其中 100 条正是 Heretic 优化所用的评估提示,因此并非留出数据):官方未改动 101 / 104,我们的 0 / 104(Q8_0),0 / 104(Q4_K_M)。
• 与官方编码器生成的图像(同一图像模型、同样种子,九条提示词)相比,PSNR 为 33.4 dB mean / 26.1 dB min over 9 prompts。没有测量被编辑的编码器在受限内容上对 图像 的改变;本仓库不包含受限内容的示例图片。
• 每张图耗时: 见英文版(只有在 GPU 空闲时的测量才会写入)。
• 没有测量: 官方编码器处理较差的提示上被编辑编码器的表现、参考图像编辑、其他语言、其他硬件上的速度。

许可与致谢

Apache-2.0(与原始 Qwen3-VL-4B-Instruct 相同);文件经过修改(消融并量化),见 NOTICE。属于 Sakura Media 系列。致谢:Qwen / Alibaba(Qwen3-VL-4B-Instruct)、Black Forest Labs、p-e-w/heretic(AGPL-3.0,仅作为工具使用,不随本仓库分发)、ggml-org/llama.cpp、leejet/stable-diffusion.cpp、city96/ComfyUI-GGUF。