AI Transparent Background Image Generator (Qwen Image 2.1)

Yes — Qwen Image 2.1 on VdoBloom can generate PNG or WebP images with a genuine alpha channel by setting background: transparent, as long as your prompt describes only the subject and not a background, scene, or surface.

Disclosure: VdoBloom is our platform. Capabilities below are taken from the model’s API specification; we have not published our own test results yet.

What “transparent background” actually means here

A lot of “AI transparent background” tools are really doing two separate steps behind the scenes: generate an image on some background, then run background removal on top of it. That second step is a segmentation guess, and it shows up as haloing, chopped hair strands, or soft edges that don’t hold up when you drop the cutout onto a colored layout.

Qwen Image 2.1’s background: transparent option works differently — per the model’s API spec, it renders the subject directly onto an alpha channel as part of generation, and outputs a PNG or WebP file that actually carries that channel. There’s no separate cutout pass to get wrong, because there was never an opaque background to remove.

The one rule that breaks transparency

This is the detail most people miss, and it’s explicit in the API spec: when background: transparent is set, your prompt should describe only the subject. If your prompt mentions a background, scene, surface, floor, sky, or any environment at all, the model may render that environment instead of leaving it transparent — because you asked it to.

• Do: “A red vintage motorcycle, three-quarter angle, studio lighting on the subject.”
• Avoid: “A red vintage motorcycle parked on a cobblestone street at sunset.” — the street and sunset are background elements and can defeat transparency.

There’s a second setting that interacts with this: enhance_prompt is on by default and rewrites your prompt into a fuller scene description. Since a fuller scene is exactly what you don’t want for a transparent cutout, the spec’s own guidance points toward turning enhance_prompt off when you’re generating a transparent subject — we haven’t run controlled tests to confirm how often it matters, but it’s the safer default for this use case.

Where to find it on VdoBloom

Qwen Image 2.1 shows up in two places depending on whether you’re starting from scratch or editing an existing image:

1. Text to Image (qwen2-1/text-to-image) at /dashboard/images/generate/ — generate a new subject on a transparent background from a text prompt.
2. Image Edit (qwen2-1/image-to-image) at /dashboard/images/edit/ — use 1–10 reference images to guide a new transparent-background render.

Formats: PNG, WebP, and why not JPEG

Output format is a separate setting from transparency, and the two interact in one important way — JPEG has no alpha channel, so it physically cannot store transparency. If you need a see-through PNG or WebP, JPEG is off the table for that generation.

Format	Supports transparent background	Notes
PNG	Yes (real alpha channel)	Default output format; widest compatibility for logos, stickers, overlays
WebP	Yes (real alpha channel)	Smaller file size than PNG at similar quality
JPEG	No	Cannot be combined with background: transparent

Resolution and aspect ratio

Qwen Image 2.1 offers nine aspect ratios — 1:1, 4:3, 3:4, 3:2, 2:3, 16:9, 9:16, 21:9, and 9:21 — each available at two resolution tiers, 1K and 2K. 2K takes longer to render. Here are the pixel dimensions for the most commonly used ratios:

Aspect ratio	1K	2K	Typical use
1:1	1024 × 1024	2048 × 2048	Icons, product thumbnails, social avatars
16:9	1376 × 768	2752 × 1536	Thumbnail overlays, banner cutouts
9:16	768 × 1376	1536 × 2752	Story/reel overlay graphics
21:9	1568 × 672	3136 × 1344	Wide banner assets, UI headers

The remaining ratios (4:3, 3:4, 3:2, 2:3, 9:21) are available in the same 1K/2K tiers for cases in between — pick whichever matches the canvas you’re dropping the cutout into.

Step-by-step: generate a transparent PNG

1. Go to /dashboard/images/generate/ and select the Text to Image tab.
2. Choose Qwen Image 2.1 as the model.
3. Write a subject-only prompt — no floor, wall, sky, or setting.
4. Set output format to PNG (or WebP) and enable background: transparent.
5. Turn off enhance_prompt so the rewrite doesn’t add a scene back in.
6. Pick your aspect ratio and resolution (1K for speed, 2K for print-quality or heavy cropping).
7. Optionally set a seed if you want to reproduce this exact result later with small prompt tweaks.
8. Generate. The credit cost is shown before you confirm: 3 credits per 1K image and 5 credits per 2K image, for both Text to Image and Image Edit. New accounts start with 10 free credits, so a first transparent PNG costs nothing.

Editing an existing image with a transparent output

The Image Edit tab (qwen2-1/image-to-image) accepts 1–10 reference images and can also output with a transparent background, using the same subject-only-prompt rule above. It also supports a separate mask_url inpainting mode — white areas of the mask get changed, black areas are kept, and it requires exactly one reference image at the same aspect ratio as the mask. One important limitation: mask-based inpainting cannot be combined with background: transparent in the same request — they’re mutually exclusive modes, so choose one or the other depending on whether you’re replacing content in place (mask) or extracting a clean cutout (transparent).

What people actually use this for

• Stickers — subject-only PNGs ready for messaging apps or print-on-demand sticker sheets.
• Logos and icons — clean alpha edges for brand marks that need to sit on any color.
• Product cutouts — marketplace listings where the platform expects a transparent or white product shot.
• Video thumbnail overlays — a subject rendered with no background so it composites over a video frame.
• Game and UI assets — icons, buttons, and character sprites that need to layer over other art.
• Print-on-demand graphics — designs meant to sit on a shirt, mug, or case color rather than their own background.

Quick troubleshooting

Problem	Likely cause	Fix
Background isn’t actually transparent	Prompt mentioned a scene, surface, or setting	Rewrite the prompt to describe only the subject
Background isn’t transparent even with a clean prompt	enhance_prompt rewrote it into a fuller scene	Turn enhance_prompt off
Alpha channel missing in the downloaded file	Output format was set to JPEG	Switch to PNG or WebP before generating
Can’t combine masking with transparency	These are separate, mutually exclusive modes	Use mask_url for in-place edits, or background: transparent for a clean cutout — not both

The short version

Qwen Image 2.1 on VdoBloom generates real alpha-channel PNGs and WebPs directly — no separate background-removal step. The only thing to get right is prompt discipline: describe the subject, not the scene, and turn off enhance_prompt so nothing adds a background back in behind your back. Everything else — nine aspect ratios, 1K/2K resolution, up to ten reference images for edits, and reproducible seeds — is standard generation tooling layered on top of that one rule.