Tutorial7 min readSeptember 13, 2026

How to Use Qwen2 Image Edit: 6 Credits per Edit, 8 Ratios, Step by Step

A practical walkthrough of Qwen2 Image Edit on VdoBloom: real credit cost, the eight aspect ratios, what photos work, prompting rules, and when the Qwen 3 editors beat it.

Qwen2 Image Edit is VdoBloom’s speed-first instruction editor: 6 credits per edit (about $0.30), eight aspect ratios including 21:9 ultrawide, PNG or JPEG output, and a seed field for repeatable results. You upload one image, describe the change in plain language, and it returns a revised version — no masks, no layers, no selection tools. It is built for the iterate-and-adjust loop, not for one perfect shot.

What it actually does

Qwen2 Image Edit comes from Qwen, Alibaba’s open model family spanning language, vision and image generation. The image-editing branch is known for solid instruction-following on modest infrastructure, which is exactly the trade this model makes on VdoBloom: it is cheaper and quicker than the heavyweight editors, at a fidelity that is good rather than best-in-class. Users rate it 4.0.

Be honest with yourself about what an instruction editor is. It does not surgically patch pixels the way a masked erase does — it re-renders the image under your instruction. Small, well-scoped asks (change the jacket to red, remove the car on the left, put the sky at golden hour) hold the rest of the frame convincingly. Sweeping asks (make it cinematic and add three people) rebuild so much of the picture that the original stops being recognisable. That is a property of the technique, not a flaw in this particular model.

Real specs and credit cost

The model is flat-priced: one number per run, no resolution tiers. Dollar figures below convert at the Lite rate of $15 for 300 credits, so 1 credit is $0.05.

EditorCredits per editCost at $0.05/creditResolution controlAspect ratiosReference images
Qwen2 Image Edit6$0.30None8 (16:9 default, 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 21:9)1
Qwen Image 3.0 Edit5$0.251K / 2K8 (incl. 21:9)Multi-image
Qwen Image 3.0 Pro Edit (1K)6$0.301K8 (incl. 21:9)Multi-image
Qwen Image 3.0 Pro Edit (2K)7$0.352K8 (incl. 21:9)Multi-image
Qwen Image 2.010$0.50None7 (no 21:9)Up to 3
Qwen Image 2.0 Pro12$0.60None7 (no 21:9)Up to 3

That table contains an awkward fact and it is better said out loud: Qwen Image 3.0 Edit is cheaper at 5 credits and adds a 1K/2K resolution choice plus multi-image input. For most editing jobs it is the better default. Qwen2 Image Edit stays useful for its own reasons — a simpler control surface with an explicit output-format toggle, a seed field, and a price that is still a third of what the 2.0 Pro tier charges — but if you came here expecting the cheapest Qwen editor, the 3.0 line is it. Compare the whole family on the Qwen2 Image Edit model page and its 2.0 Pro sibling before you settle.

Which photo works best

  • Sharp and well-lit. The model re-renders what it can see; a soft, noisy phone shot in low light gives it less to preserve, and the output drifts further from the original.
  • One clear subject. Busy group scenes multiply the things that can shift between runs.
  • Framed close to your target ratio. The ratio you pick shapes the output, so starting near it avoids the model inventing content at the edges.
  • Not tiny. There is no upscaler built into this model. If your source is small, run it through the image upscaler first, then edit.

If the photo shows a real, identifiable person, you need that person’s consent before generating edited images of them — that applies to friends and colleagues as much as to public figures.

The exact steps

  1. Open AI image edit in the dashboard.
  2. Pick Qwen2 Image Edit from the model list. It shows the fast icon and a HOT content label — VdoBloom’s flexible filtering tier, which is why swimwear, dance, fitness and fashion edits that stricter providers refuse tend to go through here.
  3. Upload your source image.
  4. Choose the aspect ratio. The interface defaults to 16:9 because that is the model’s own spec default; change it to 9:16 for stories, 1:1 for feed posts, or 21:9 for a site banner.
  5. Set the output format — PNG for anything you will composite or need crisp edges on, JPEG for smaller files headed straight to the web.
  6. Write one instruction, not three. See the prompting notes below.
  7. Generate. The cost, 6 credits, is shown before the run and deducted on success.
  8. If the result is nearly right, fix the seed and change one word in the prompt. That keeps the overall composition and isolates what you changed.

Prompting an instruction editor

Three rules cover most of it. First, name what stays as well as what changes: change the wall behind her to exposed brick, keep the pose, outfit and lighting identical. Second, one change per run — stacking three edits into one prompt makes it impossible to tell which instruction caused the part you dislike, and at 6 credits a run, three separate runs cost 18 credits and teach you far more. Third, be concrete about the target: warm late-afternoon sunlight from the left beats better lighting every time.

If you cannot find the words for a look you already have in another image, run that image through image to prompt, then lift the phrases you want into your edit instruction.

A worked example

You have a product photo of a candle on a kitchen counter and you need a 21:9 banner for a site header. Upload it, set 21:9, PNG output, and prompt: keep the candle exactly as it is — same label, same shape, same colour — and place it on a dark slate surface with soft window light from the right and a clean empty area to the left of the frame for text. That is 6 credits. If the label distorts, run again with the label wording spelled out in the prompt; if it still drifts, this is the point to move up to a typography-focused model rather than paying for more rerolls. Four attempts here is 24 credits, about $1.20 — and a new account starts with 10 free credits and no card, which covers the first run and a fix.

When to use something else

  • You need 2K output: Qwen Image 3.0 Edit or 3.0 Pro Edit — 5 to 7 credits, with an explicit resolution control.
  • You need multiple reference images: the 2.0 tier takes up to three; the 3.0 line takes more.
  • Text must render perfectly inside the image: use the typography-tuned Qwen Image 2.0 Pro at 12 credits.
  • You are creating, not correcting: start in text-to-image instead — editing your way to a new concept costs more than generating it.

Every figure here is read from VdoBloom’s live credit table, and the exact cost is displayed in the interface before each run. Plan detail is on the pricing page: subscriptions from $15/month for 300 credits, one-time packs from $2.49 for 75 credits that never expire, and watermark-free downloads with commercial rights on every paid plan.

Frequently asked questions

How much does one Qwen2 Image Edit run cost?

6 credits, flat, regardless of aspect ratio or output format — about $0.30 at the Lite rate of $15 for 300 credits.

Can it edit more than one image at a time?

No. Treat it as a one-image-in, one-image-out editor. For multi-reference editing use the Qwen 3 edit models or the Qwen Image 2.0 tier, which accepts up to three references.

Which aspect ratios does it support?

Eight: 16:9 (the default), 1:1, 2:3, 3:2, 3:4, 4:3, 9:16 and 21:9. The 21:9 ultrawide option is something the Qwen Image 2.0 tier does not offer.

Will it refuse borderline prompts?

It carries the HOT content label, VdoBloom’s most permissive tier, so stylised, swimwear, dance and fashion edits that mainstream editors bounce generally go through. Explicit and illegal content is blocked platform-wide, on every model.

Does it keep faces consistent?

Reasonably, on small scoped edits — and much better when you fix the seed and name the face as something that must stay unchanged. Large rewrites of the scene will shift it.

Ready to try it?

Create your first AI video in minutes — no credit card required.

Start Creating Free →