Qwen Image license: is Qwen-Image open source under Apache 2.0?
The Qwen-Image model card says the weights are Apache 2.0 and were released on 2025-08-04. What that covers, and how the hosted qwen ids on Sume differ.
Yes: the Qwen-Image model card on Hugging Face says "Qwen-Image is licensed under Apache 2.0", and its license label reads apache-2.0. The card says the weights were released on 2025.08.04. That covers the Qwen/Qwen-Image release; it says nothing about other Qwen image versions or about hosted services.
Facts about Qwen-Image are from its model card, read 2026-09-29. Sume facts are from Image generation and the image catalog code.
What does the Qwen-Image card list?
| Point | What the card says |
|---|---|
| License | Apache 2.0 (apache-2.0 label) |
| Weights release | 2025.08.04, on Hugging Face and ModelScope |
| Focus | An image generation foundation model with advances in complex text rendering and precise image editing |
| 1:1 and 16:9 example sizes | 1328 x 1328 and 1664 x 928 in the card's Quick Start code |
What does the card say Qwen-Image does?
The introduction says Qwen-Image has advances in complex text rendering, especially for Chinese, and in precise image editing. The showcase text lists editing operations such as style transfer, object insertion or removal, detail enhancement, text editing within images and human pose manipulation, and understanding tasks such as object detection and depth estimation.
Those are the open release's capabilities as its authors describe them. They are not claims about any hosted model.
Can I use Qwen-Image for commercial work?
The card labels the weights apache-2.0 and says Qwen-Image is licensed under Apache 2.0. Read the license file in the repository for the exact terms; this post is not legal advice and does not summarize them.
How does the card run Qwen-Image?
The card's Quick Start uses the Diffusers DiffusionPipeline with Qwen/Qwen-Image, bfloat16 on CUDA, num_inference_steps=50 and true_cfg_scale=4.0. It lists sizes for seven aspect ratios, from 1328 x 1328 for 1:1 to 1056 x 1584 for 2:3. The card gives no VRAM figure in the parts read, so this post gives none.
Does Sume run Qwen-Image?
Sume's image catalog code lists two hosted ids, qwen/qwen-image and qwen/qwen-image-max, and the API lists the image models it serves at GET /v1/images/models. The docs do not say which checkpoint sits behind a hosted Qwen id, so do not assume it is the open Qwen-Image release, and do not assume it shares the Apache license.
Check the model list for what your key can call today:
curl "https://api.sume.com/v1/images/models" \
-H "Authorization: Bearer $SUME_API_KEY"What does this post not cover?
- Newer Qwen image releases and edit variants: their cards may carry other licenses and were not read.
- Rules for generated output: the card statement is about the model license, and this post does not interpret it.
- Quality: this is a licensing answer, not a comparison.
Sources
Related posts
More in Models
- Recraft V4 SVG: can you get vector output through Sume?
Recraft says V4 can make editable SVG. Sume's recraft/recraft-v4 advertises webp output only. What that means, plus a working call.
- Remove shadow from photo with AI: edit it or cut it out
To remove a shadow from a photo with AI, send the photo to an image-edit model, name the one shadow to remove, and list what stays. Prompts, cost, limits.
- Does generate_audio change Seedance 2.0's price?
No. fal says Seedance 2.0 audio costs the same on or off, and Sume's Seedance price has no audio input: seedance-2 reserves about $0.38 per 720p 16:9 second.
- Seedance 2.0 max length: 15 seconds, and how to go longer
Seedance 2.0 clips run 4 to 15 seconds on Sume and on fal. For 16 to 30 seconds use seedance-2.5 ($17.34 at 720p) or chain clips with first_frame.
Written by Sume