Qwen Image 2.1 Edit: AI Image Editor Online

All of this is enforced by VdoBloom's Qwen Image 2.1 service and the Image Edit tab, so a setting the model cannot take is blocked before any credits are spent.

Tick "Edit only a painted area (mask)" in the Image Edit tab, upload the one image you want to change, and paint over the region. VdoBloom turns your painting into a black-and-white mask and sends it with the image; the model edits inside the mask and leaves the rest untouched. This is the same paint-the-area idea as FLUX Erase, except that Qwen Image 2.1 Edit follows a prompt for what goes in the painted area instead of only removing what was there.

Mask mode has its own rules, and they are worth knowing before you paint: it takes exactly one image; the aspect-ratio and enhance-prompt settings are ignored (the output follows your image); and it cannot be combined with a transparent background, so the transparency toggle disappears while the mask is on. Turning the mask on clears the upload slots, because the second slot becomes the mask.

Outside mask mode, tick "Transparent background" and the model returns the subject on a genuine alpha channel. That needs PNG or WebP: JPEG has no alpha channel, so VdoBloom switches JPEG to PNG when you turn transparency on, and the service rejects a transparent JPEG request before it reaches the provider. Describe only the subject ("the red sneaker, side view"), not a scene, or the model may paint a background back in.

Turning off Enhance prompt helps here, because the rewrite can add scene language to a short instruction. Transparency costs nothing extra: 3 credits at 1K, 5 at 2K. The walkthrough below covers one-photo cut-outs step by step.

Only resolution changes the price. Mask or instruction mode, one reference or ten, transparent or opaque, PNG or WebP: all the same. The dollar figures use the Lite plan rate of $0.05 a credit ($15 for 300). Failed edits refund automatically, and the cost is shown before you click.

The comparison that matters is features first, price second. Qwen Image 2.1 Edit is the only one in this table with both a painted mask and transparent output; Qwen Image 3.0 Edit and Pro Edit take the same 10 references with a much shorter prompt; Qwen2 Image Edit and Nano Banana Edit are flat-priced instruction editors; FLUX Erase is mask-only removal.

It runs in the Image Edit tab (the AI Photo Editor).

In instruction mode the model already has your images, so write the change, not the scene: "make the jacket navy blue, keep the face and background unchanged". With several references, name them by order ("the logo from image 3 on the mug in image 1"); the routing guide's note that results are best with four or fewer references is worth following when you can. Use the auto aspect ratio to keep the shape of your first image, or pick a ratio when the result is going into a fixed slot such as a 9:16 story or a 21:9 banner.

Enhance prompt is on by default and expands short instructions into fuller descriptions. That helps a vague one-liner but can drift from exact wording, so switch it off when you have written precisely what you want, and always for transparent cut-outs.

There is no 4K; 2K is the top tier. Mask mode is one image only, ignores the ratio and enhance settings, and cannot output a transparent background, so a masked edit and a cut-out are two separate runs. Transparency never works with JPEG. Results with many references can lose fidelity, and VdoBloom's own guide suggests four or fewer. Uploads are capped at 10 MB each. The model runs only through KIE, with no second provider behind it, so if KIE is down the job fails and the credits come back automatically. It is a hosted model, not a download.

VdoBloom does not publish quality scores for models. At 3 credits a 1K edit, a test run on your own photo is the quickest way to see whether the result suits the job.