GUIDE / CREATE

Image generation: prompts, parameters and actual examples.

Go from a description to a downloadable image. Follow two actual platform examples—an anime illustration and an e-commerce poster—through model selection, prompting, generation and cost checks.

ANIME

Anime illustration prompt

Landscape 16:9: combine a character, seaside town and sunset into a complete illustration.

Learn more

ECOMMERCE

E-commerce poster prompt

Portrait 4:5: specify products, text areas, materials and Chinese copy.

Learn more

PRICE

Check current prices

These recorded runs charged 0.10 bullets per image, equivalent to CNY 0.10. Future charges depend on the current model, parameters and billing rules.

Learn more

Specify five things in your prompt

Subject, scene, composition, lighting and purpose are easier to adjust than a stack of adjectives. Copy the original prompts below, or replace the subject with your own idea.

  • Subject: who or which product, and what it should look like. Do not substitute an existing anime character for an original one.
  • Scene and lighting: a seaside platform at sunset, or a cream display stand with soft studio lighting.
  • Composition: a character in the right third, or a centered product with the headline above.
  • Purpose: landscape wallpaper or an online-store poster. Leave room for text and cropping.
  • Constraints: no watermarks, logos, buy buttons or prices you did not supply. Give exact wording if you need text.

What do 1K, 2K, 4K and aspect ratio each control?

1K, 2K and 4K are workspace resolution tiers. Values such as 16:9, 9:16 and 4:5 are aspect ratios. Choose the tier for resolution and its associated price, and the ratio for composition. Actual output dimensions also depend on the model; a tier does not define one pixel size for every image.

Check the intended use, pixels and cost in order
DecisionWhere to lookHow to choose or verify
Resolution tierThe workspace model and resolution controlsCheck the tiers available for the model, then choose for your final use and budget. Do not assume every model supports 4K.
Aspect ratioThe workspace ratio controlsConsider 16:9 for a landscape wallpaper or 4:5 for a portrait poster. Leave room for cropping and text, and check model support.
Price before generatingThe public models and prices pageCheck the current quote for the same model, tier and relevant parameters. A guide's historical charge is not a current price.
Actual pixelsThe downloaded image's propertiesCheck width × height against your display or layout requirements instead of relying only on the 1K, 2K or 4K label.
Final chargeThe usage log at the generation timeMatch the model, count, dimensions, quality and actual charge. Regenerating creates a new request and charge.
  • Both verified examples on this page used 1K. They do not establish a comparison of 2K/4K output quality or generation time.
  • Auto and a manual aspect ratio may produce different composition choices. Choose a supported fixed ratio for important layouts, then verify the downloaded dimensions.

How do Nano Banana parameters differ from the GPT Image examples?

Switching models requires more than changing model. Nano Banana 2, 2.1 and Pro use POST /v1/images/generations, with their own resolution, ratio and editing constraints. Both historical images below were generated with GPT Image 2.5; they are not Nano Banana test outputs.

Check these parameters before reusing a prompt with Nano Banana
Parameter or operationCurrent Nano Banana integrationWhen adapting a GPT example
modelnano-banana-2 / nano-banana-2.1 / nano-banana-proChoose an available version from the current catalog. The workspace family name is not an API model ID.
size1K / 2K / 4KUse native tiers. GPT's 2880x2880 is not Nano's 4K tier.
ratio1:1, 16:9, 9:16, 4:3, 3:4; omitted or Auto uses 1:1The landscape example can use 16:9. Adapt the 4:5 poster to a supported portrait ratio and revise the composition.
quality and nquality=auto, n=1Generate one image per request. Do not copy another model's quality tiers or batch count.
Reference-image editingNot supported by the current Nano integrationThe examples do not upload a reference image. Choose an explicitly compatible model when you need editing.
Pixels and costVerify the returned image and usage recordCheck actual dimensions even with native tiers. The historical GPT charges do not apply to Nano.

1. Choose the model and parameters

Sign in and open Generate. Both examples used GPT Image 2.5 (model ID: gpt-image-2.5) at the 1K tier. Check the current price in Models & prices, then choose an aspect ratio for the image's purpose.

  • Anime example: 1K, 16:9; actual output 1536 × 864 pixels.
  • E-commerce example: 1K, 4:5; actual output 1024 × 1280 pixels.
  • 1K is a resolution tier, not a promise that every ratio outputs 1024 × 1024. Download the image to check its actual dimensions.
  • The current workspace generates one image per request with auto quality. Neither example used a reference image.

2. Paste the prompt and wait

Paste the example into Describe the image, check the model, 1K tier and ratio, then select Generate image. When the canvas shows Generated · saved to library, inspect the subject, composition, text and details.

  • The anime request took about 55 seconds and the poster about 27 seconds. These are recorded runs, not guaranteed response times.
  • Save the current image before changing one important condition. Another generation creates a new request and charge.
  • If generation fails, read the error first. Model, parameters, balance and safety checks can affect a request. Do not keep clicking repeatedly.

3. Download and check the charge

Select Download to save the image to your device. In Logs, find the successful record by model and time, then check image count, dimensions, quality and cost.

  • On 2026-10-11, each successful request charged 0.10 bullets (CNY 0.10), totaling CNY 0.20. Current billing rules determine future charges.
  • Keep the prompt, model, ratio and resolution. The image library does not retain a complete set of your creative notes.
  • Hosted generated images expire. Download important work promptly. These examples use downloaded static copies.

What does one image cost, and how do I verify its size?

These two platform runs from 11 October 2026 show how tier, ratio, output pixels and charge relate. Prices can change with the model, tier, quality and billing rules. The records are not a current universal quote or a guarantee of identical output dimensions.

2026-10-11: two recorded GPT Image 2.5 runs at 1K
ExampleWorkspace parametersOriginal output pixelsLogged charge
Anime illustration1K · 16:9 · auto · 1 image1536 × 8640.10 bullets, equivalent to CNY 0.10
E-commerce poster1K · 4:5 · auto · 1 image1024 × 12800.10 bullets, equivalent to CNY 0.10
  • Both images used 1K, but their ratios and actual dimensions differed. The examples below include complete prompts, outputs and parameters.
  • Check current models and prices, generate, inspect the downloaded dimensions, then verify the logged charge. Do not infer another model's cost from image area.

Adjust what you can see

Adjusting an actual result is easier than adding many conflicting requirements. Keep the model and parameters fixed, then change one subject, position, lighting or text condition at a time.

  • Character too small: specify a half-body composition, position and crop.
  • Product lacks emphasis: reduce background decoration and specify product count, placement and soft side lighting.
  • Text is inaccurate: check every character. If needed, generate a text-free base and use a design tool for layout, noting the post-processing.
  • Real product promotion requires accurate appearance, branding and copy. This fictional skincare example does not establish an accurate depiction of a real product.

MADE WITH AK47TOKEN

Two images, from prompts to actual outputs.

These images were generated in AK47Token on 2026-10-11. Each is an actual individual output, resized and compressed only for web publishing, without retouching or rearranging text. Models are stochastic; the same prompt and parameters do not guarantee the same image.

REAL GENERATION / 2026.10.11

Anime illustration: seaside platform at sunset

A landscape illustration with a clear subject and layered background. Specify position, clothing and lighting, and use 16:9 to leave room for the sea and town.

Actual platform output: an adult traveler in a white coat and blue skirt on a seaside platform, with a train, town and golden sunset behind her
Actual platform output · open full image

Recorded generation

Model ID
gpt-image-2.5
Resolution tier / ratio
1K / 16:9
Actual pixels
1536 × 864
Quality / count
auto / 1 image
Recorded charge
0.10 bullets (CNY 0.10)
Request duration
54.50 s

This is the recorded charge, not a permanently fixed price. Before generating again, check current model prices。

Copy this prompt

Open the image workspace, select the recorded model, resolution tier and ratio, paste the original Chinese prompt below and select Generate image.

原创二次元插画,横版16:9。主体是一位明确成年的女性旅行者,穿白色外套和蓝色长裙,背小肩包,站在海边铁路站台,人物放在画面右侧三分之一处,脸和上半身完整可见,衣着完整得体。傍晚的金色侧光,海风吹起头发,远处有小列车、海边小镇、蓝绿色海面和粉橙色积云。清晰细腻的线稿,手绘动画背景,前中后景有空间层次,人物比例自然,整体明亮温暖,适合横版壁纸。不要任何文字、标志、水印,不使用已有动漫角色。

Inspect the result, then adjust

  • The subject appears on the right, with the sea, town and train visible. Sunset and hair highlights form consistent lighting.
  • The lower body is not fully shown. For a full-body design, replace the original instruction about the face and upper body with a request for the entire body, complete feet and margin, then test separately.
  • Inspect the train, platform and other details at full size. Projects requiring precise structures cannot be accepted from a thumbnail alone.
Try it in the workspace

REAL GENERATION / 2026.10.11

E-commerce poster: fictional skincare products

A complete promotional poster with products, Chinese headlines and texture details. Specify text and product areas, then provide exact copy so the result is more than a product still life.

Actual platform poster: fictional skincare bottles and cream below a Chinese summer-skincare headline, with two texture-detail circles at the bottom
Actual platform output · open full image

Recorded generation

Model ID
gpt-image-2.5
Resolution tier / ratio
1K / 4:5
Actual pixels
1024 × 1280
Quality / count
auto / 1 image
Recorded charge
0.10 bullets (CNY 0.10)
Request duration
27.25 s

This is the recorded charge, not a permanently fixed price. Before generating again, check current model prices。

Copy this prompt

Open the image workspace, select the recorded model, resolution tier and ratio, paste the original Chinese prompt below and select Generate image.

生成一张完整的原创电商护肤品宣传海报,竖版4:5,画面应让人一眼看出是商品宣传设计,而不只是静物照片。虚构的无品牌护肤品套装:一只透明蜜桃色滴管精华瓶、一只乳白色面霜罐、一只磨砂小瓶,产品完整摆在米白色展示台上,占画面下部中央。暖米色和浅蜜桃色背景,柔和棚拍侧光,细腻真实的玻璃、奶油和水滴质感。上方留出整齐的文字区,准确排版简体中文大标题“夏日护肤季”,副标题“清爽水润 · 轻盈好吸收”,小标签“新品上市”。下方放两枚精华滴液、面霜质地的圆形细节图,标签分别为“水润精华”“细腻乳霜”。字体清楚、层级分明、留白充足,像可用于网店首页的成品活动海报。不添加立即选购按钮、价格、折扣、品牌名、销量、功效承诺、认证徽章、水印或额外文字。

Inspect the result, then adjust

  • The headline, subtitle and new-product label are present. The product area and two texture details are distinct, with no buy button or price.
  • Check each character beneath the texture circles in the full image. A real campaign needs human review of product details and wording; one successful text result is not a guarantee for every model.
  • This is a fictional product example, not a real brand photograph, customer advertisement or endorsement. Check appearance and promotional claims when using a real product.
Try it in the workspace

FAQ

Questions & answers

Are longer prompts always better?

No. Repeated or conflicting instructions can pull a result away from your goal. Clear subjects and composition are usually easier to adjust than many style adjectives.

Will copying the prompt produce the exact image?

No. This guide reproduces the model, parameters and workflow, not an identical random output. It shows two individual actual outputs, not a composite of selected best results.

Why is a 1K image not 1024 × 1024 in the log?

Resolution tier and aspect ratio are separate settings. The recorded 16:9 output is 1536 × 864 and the 4:5 output is 1024 × 1280; both use their corresponding 1K billing tier.

Will every image cost CNY 0.10?

Not guaranteed. These runs on 2026-10-11 charged 0.10 bullets per image, equivalent to CNY 0.10 at CNY 1 per bullet. Model, tier, quality and price changes may affect future costs. Check the current price before generating and verify the log afterward.

Can I upload a product reference in this workspace?

This guide verifies text-to-image generation. The demonstrated interface has no reference-upload step. Check model and endpoint support separately for reference editing; a photo alone does not guarantee a matching appearance.

Ready to get started?