Qwen-Image-2512.
Qwen-Image-2512 is an open-source image model from Alibaba (Qwen team). Windpaint doesn't serve it yet, so there's no model id or price. Request early access to start with the live models.
- Status
- Coming soon
- Task
- Text to image
- Input
- Prompt
- Output
- Images
- Developer
- Alibaba (Qwen team)
- Parameters
- 20B
- Licence
- Apache-2.0
- Released
- December 2025
What the model does.
Paraphrased from the official model card.
Qwen-Image is a 20B text-to-image model from Alibaba's Qwen team. It's built for putting legible text into images, including long passages and Chinese as well as English.
Qwen-Image-2512 is the December 2025 update of the model first released in August 2025. Its card lists three changes: people look less artificial, natural detail such as fur and landscapes is finer, and text and layout render more accurately.
Coming soon to Windpaint.
Qwen-Image-2512 isn't served, so it has no model id, no price and no request example here. When it goes live, this page gets its tiers, prices and request fields like the live models have.
Want this model? Request early access. We're adding models as we add capacity.
Live for images today: Z-Image Turbo (text to image) and FLUX.2 [klein] 4B (text to image). See Image generation for what the API does now.
Is Qwen-Image-2512 licensed for commercial use?
Licence: Apache-2.0.
Yes. Apache-2.0 allows commercial use, including running the model behind a paid API, with no revenue or user caps.
- Apache-2.0, per the card metadata and the repo's License Agreement section ("Qwen-Image is licensed under Apache 2.0"). Qwen-Image repo
This is our reading of the licence, not legal advice. Read the licence text before you rely on it.
Qwen-Image-2512 hardware requirements.
Published by Alibaba (Qwen team) for running the weights yourself. These are the developer's figures, not Windpaint's.
- Published VRAM
- Not stated
- Applies to
- No VRAM figure on the card. bf16 on CUDA; the repo ships a multi-GPU server.
- The card's code loads the model in bfloat16 on CUDA and falls back to float32 on CPU. The card gives no VRAM figure. Qwen-Image-2512 model card
- The Qwen-Image repo ships a multi-GPU Gradio API server, configured with
NUM_GPUS_TO_USE(example: 4),TASK_QUEUE_SIZEandTASK_TIMEOUT. Qwen-Image repo - The official README's community section says DiffSynth-Studio (from ModelScope, a community tool) supports Qwen-Image with layer-by-layer offload (inference within 4GB VRAM), FP8 quantization, and LoRA and full training. The README doesn't say which checkpoints this was tested on. Qwen-Image repo
Qwen-Image-2512 native resolution.
What the developer's own code and examples use.
- Card aspect-ratio presets: 1:1 1328 × 1328, 16:9 1664 × 928, 9:16 928 × 1664, 4:3 1472 × 1104, 3:4 1104 × 1472, 3:2 1584 × 1056, 2:3 1056 × 1584. Qwen-Image-2512 model card
Recommended settings for Qwen-Image-2512.
Steps, guidance and defaults from the developer's card and code, for running the weights yourself.
- The card's example uses
num_inference_steps=50andtrue_cfg_scale=4.0, with a negative prompt. Qwen-Image-2512 model card
Qwen-Image-2512 prompt guide.
From Alibaba (Qwen team)'s own prompt guide, prompt tooling and examples.
- The repo recommends the 2512 prompt-enhancing tool (
src/), which rewrites prompts with a Qwen LLM (qwen-plus via DashScope). Qwen-Image repoexamples/ tools/ prompt_utils_2512.py - The official rewrite guidance puts every piece of visible text in English double quotes and describes its content, position, layout direction, font style, colour, size and presentation (printed, embroidered, neon). It never changes proper nouns, and it states "The image contains no recognizable text." when there is none. Qwen-Image repo, prompt_utils_2512.py
- The same guidance asks for the overall artistic style to be stated (realistic photography, anime illustration, movie poster, 3D rendering and so on), and for continuous prose instead of Markdown lists or headings. Qwen-Image repo, prompt_utils_2512.py
- The card's example uses a Chinese negative prompt against low resolution, low quality, malformed limbs and fingers, oversaturation, a waxy look, faces without detail, over-smoothing, an "AI look", messy composition, and blurry or distorted text. Qwen-Image-2512 model card
Questions.
Can I call Qwen-Image-2512 on Windpaint today?
No. Qwen-Image-2512 isn't served yet, so it has no model id, no price and no entry in the catalog endpoint. For images today, z-image-turbo and flux-2-klein-4b are live. There's no date for Qwen-Image-2512.
What resolutions does Qwen-Image-2512 support?
The card lists seven presets: 1328x1328 (1:1), 1664x928 (16:9), 928x1664 (9:16), 1472x1104 (4:3), 1104x1472 (3:4), 1584x1056 (3:2) and 1056x1584 (2:3).
What licence is Qwen-Image-2512 under?
Apache-2.0, which allows commercial use.
Does Qwen-Image-2512 use a negative prompt?
Yes. The card's example passes a negative prompt with true_cfg_scale 4.0 and 50 steps.
Is there a prompt guide for Qwen-Image-2512?
The Qwen team publishes a 2512 prompt-rewriting tool. It puts all text to render in double quotes, states the art style, and keeps proper nouns unchanged.
Is Qwen-Image-2512 open source?
Yes. Its weights are licensed Apache-2.0, an OSI-approved open-source licence.
How do I ask for Qwen-Image-2512?
Request early access at app.windpaint.ai/register. We're letting people in gradually while we add models and capacity.