Qwen-ImageImage Mini and Image Fast

Make first frames, product stills and reference edits with Alibaba's open Qwen-Image models, then animate any image into a Seedance clip.

Real renders not open yetLicenseApache-2.0Model weights on Hugging Face ↗For AI agents

Quoted before you render
Refunded on failure
Commercial use on paid renders

Image Mini · 1K1 cr≈ $0.01 · ≈ 1,000 per $10

Image Mini · 2K3 cr≈ $0.03 · ≈ 333 per $10

Image Fast · 1K2 cr≈ $0.02 · ≈ 500 per $10

Image Fast · 2K6 cr≈ $0.06 · ≈ 166 per $10

Editing costs a little more: 2 cr on Image Mini and 3 cr on Image Fast at 1K.

Animate any image into a Seedance clip from 8 cr (Mini · 480p · 4 s).

Not open yetPrefilled example

FIRST FRAME

Prompt: Gold watch on black glass with a mirror reflection, soft specular highlights, dark studio, square product still.

Image Fast
Qwen-Image-2512
1K
1:1

Quote2 cr

≈ $0.02 · Refunded if it fails

Real renders are not open yet

Leave your email and we will write once, when they open.

Exampleseach with its receipt

Every example lists its source, tier, size and cost. The frames are stock photos, not Qwen-Image output.

Still 01

Product still

Black-and-white still: a woman sprays perfume toward her neck, hard side light, a fine visible mist, deep shadows.

Source: Stock photo
Spec: Image Fast · 1K · generate
Cost: 2 cr ≈ $0.02

Draft 02

Storyboard frame

Close-up in a café mirror, warm practical lights, she glances back over her shoulder. Rough frame for a drama shot.

Source: Stock photo
Spec: Image Mini · 1K · generate
Cost: 1 cr ≈ $0.01

Edit 03

Background swap

Keep the sneaker and its wooden box exactly as they are; darken the background and add a soft top light.

Source: Stock photo
Spec: Image Fast · 2K · edit
Cost: 10 cr ≈ $0.10

Two tiersone page

The same open-weight models, run two ways. Pick by the job: drafts or production.

Image Mini and Image Fast compared
Feature	Image Mini	Image Fast
Role	Economy · draft	Recommended
Generate	Qwen-Image-2512 · Lightning 4-step`qwen-image-2512-lightning`	Qwen-Image-2512`qwen-image-2512`
Edit	Qwen-Image-Edit-2511 · Lightning`qwen-image-edit-2511-lightning`	Qwen-Image-Edit-2511`qwen-image-edit-2511`
1K image	1 cr≈ $0.01Edit 2 cr	2 cr≈ $0.02Edit 3 cr
2K image	3 cr≈ $0.03Edit 4 cr	6 cr≈ $0.06Edit 10 cr
Best for	Quick candidates, storyboard drafts, batch idea tests	Final first frames, reference edits, text and fine detail
License	Apache-2.0	Apache-2.0

Image Fast is the production tier, not a speed promise. Lightning is a distilled 4-step adapter published by lightx2v; it is not a model released by Alibaba's Qwen team.

What itdoes well

What these models are good at, and how they feed the video engines.

Text that reads

Qwen-Image is strong at legible text in English and Chinese: labels, signs, posters. Put the exact words in quotes.

Edit with a sentence

The edit models take your image plus an instruction: swap a background, change a color, remove an object.

Keep what you did not mention

Edits aim to leave the rest of the frame alone, so a label or silhouette stays recognizable. Check fine detail before you animate.

First frames for video

Every image gets an Animate button that hands it to Seedance as the first frame, from 8 cr per clip.

How to writea prompt that lands

Describe the frame like a photographer: subject, composition, light. Quote any text you want to appear.

Three layers

Gold watch on black glass, square frame, low angle, macro lens feel, soft specular highlights, mirror reflection.

Subject · the product and what surrounds it.
Composition · frame, angle, lens feel.
Light and finish · light direction and texture.

Quote the text

Exact words in quotes render more reliably than a description of them.

A shop sign that reads "OPEN LATE" in white neon.

Pick the ratio for the job

Make a first frame in the ratio the video will use, so nothing gets cropped later.

9:16 for Reels and TikTok, 16:9 for YouTube.

For edits, say what stays

Name what changes and what must not.

Replace the background with a beach at dusk; keep the bottle and label unchanged.

Parametersand limits

Model IDs (API): qwen-image-2512-lightningImage Mini · generateqwen-image-edit-2511-lightningImage Mini · editqwen-image-2512Image Fast · generateqwen-image-edit-2511Image Fast · edit
Output size: 1K · 1344×768 · 768×1344 · 1024×10242K · 2048×1152 · 1152×2048 · 2048×204816:9 · 9:16 · 1:1
Inputs: Generate models take a text prompt. Edit models also need image_url, and generate models reject it.
Weights: Qwen/Qwen-Image-2512 lightx2v/Qwen-Image-2512-Lightning Qwen/Qwen-Image-Edit-2511 lightx2v/Qwen-Image-Edit-2511-Lightning
License: Apache-2.0 for both models and both Lightning adapters.

Billingin detail

Per image, by tier, size and operation. Editing costs a little more than generating, because the edit model does more work per image.

Qwen image price by tier and size
Tier	1K	2K
Image MiniGenerate	1 cr≈ $0.01	3 cr≈ $0.03
Image MiniEdit	2 cr≈ $0.02	4 cr≈ $0.04
Image FastGenerate	2 cr≈ $0.02	6 cr≈ $0.06
Image FastEdit	3 cr≈ $0.03	10 cr≈ $0.10

Failed images: The held credits are refunded; nothing is charged.
Free tries: Free Image Mini tries, when offered, carry a watermark and are for non-commercial use. Paid images carry no watermark and can be used commercially.
Animate: Priced as a Seedance clip, from 8 cr (Seedance 2.0 Mini · 480p · 4 s).

APIquick start

The same async flow as video: quote cap, idempotency key, job id.

Request API access in the studio; once it is open, create a key. It is shown once.
POST to /v1/images/generations with a model id, size and aspect ratio.
Edit models also take image_url; generate models reject it.
Get 202 and a job id; poll it or pass webhook_url.

Docs llms.txt

For AI agents: Plain text for AI agents: inputs, legal values, every price and a request exampleqwen-image-2512-lightningqwen-image-edit-2511-lightningqwen-image-2512qwen-image-edit-2511

Real renders are not open yet.

Request

curl https://api.pershot.app/v1/images/generations \
  -H "Authorization: Bearer $API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen-image-2512",
    "prompt": "Gold watch on black glass with a mirror reflection, soft specular highlights, dark studio, square product still.",
    "size": "1K",
    "aspect_ratio": "1:1",
    "max_credits": 2
  }'

Response

{
  "id": "job_01J9ZM2Q5C",
  "object": "job",
  "kind": "image",
  "status": "queued",
  "money_status": "reserved",
  "requested_model": "qwen-image-2512",
  "credits_quoted": 2,
  "credits_charged": null
}

Live channelsas measured

One model can run on more than one supplier. Each row is one channel: its state right now, and its success rate and median render time over the last 24 hours. Refreshed every minute.

Live channels for Qwen image
Channel	Status	Success rate, 24 h	Median time, 24 h
Image Mini · generate`qwen-image-2512-lightning`
Loading live data…
Image Mini · edit`qwen-image-edit-2511-lightning`
Loading live data…
Image Fast · generate`qwen-image-2512`
Loading live data…
Image Fast · edit`qwen-image-edit-2511`
Loading live data…

Your price is the same on every channel. We switch to the backup only when the primary is unavailable or your API routing preference says so.

Measured from finished jobs in the last 24 hours. Failures caused by the request itself (a refused prompt, invalid settings, a cancellation) do not count against a channel. A channel shows numbers once it has 20 runs; until then it says not enough data yet. Time runs from the first submit to a successful result. In the API, set routing_preference to stability (the default) or speed. Platform status

Questionsanswered

How much is one image?

Image Mini costs 1 credits at 1K and 3 at 2K. Image Fast costs 2 credits at 1K and 6 at 2K. At 1K on Image Mini, $10 makes about 1,000 images.

Does editing cost more than generating?

A little. The edit model does more work per image. At 1K, Image Mini costs 1 credits to generate and 2 to edit; Image Fast costs 2 to generate and 3 to edit.

Is Image Fast faster than Image Mini?

Not necessarily. Image Fast is the production tier with standard inference; Image Mini uses a 4-step Lightning adapter for cheap drafts. Pick by the job, not the name.

Which tier should I use?

Image Mini for candidates, storyboards and idea tests. Image Fast for the first frame you will animate, reference edits, text and fine detail.

Can I use the images commercially?

Paid images, yes, under our Terms. The models and adapters are Apache-2.0. Free watermarked tries, when offered, are for non-commercial use only. Commercial use, in plain words

Can I turn an image into a video?

Yes. Every image has an Animate button that sends it to Seedance as the first frame, from 8 credits per clip.