GPT-image-2 AI Image API
GPT Image 2 is OpenAI's image-generation model. OpenAI documents image generation and editing with flexible sizes; Flatkey exposes the image routes and request controls shown below.
Output
Preview
Request summary
- Endpoint
- POST /v1/images/generations
- Model ID
- GPT-image-2
- Aspect ratio
- 2:3
- Resolution
- 1024x1024
- Quality
- high
- Outputs
- 1
Preview only. Sign in to submit this request to Flatkey.
{
"model": "gpt-image-2",
"prompt": "面部身份:\n使用附带的面部身份作为唯一面部参考。四张照片始终匹配相同的五官比例、肤色、年龄感与整体身份。\n\n身体:\n{argument name=\"人物年龄\" default=\"20岁出头的成年女性\"}。使用附带的身体参考作为唯一身体比例参考,保持真实解剖结构、自然曲线与皮肤纹理。\n主题:\n{argument name=\"主题场景\" default=\"盛唐娱乐四景\"}:曲江斗花、长安酒肆、乐坊舞毕、马球赛后。\n\n主体:\n分别生成四张独立的纵向2:3彩色写实照片,不要拼图。四张保持同一人物,但根据活动更换符合盛唐审美的发型、服装、表情和摄影语言。\n\n第一张「花隙里的胜负揭晓」:\n曲江斗花会刚揭晓结果。镜头隔着牡丹与彩幄拍摄人物大特写。她梳盛唐半翻髻,佩戴不对称簪花,穿浅杏白、石青与牡丹红组成的现代审美唐风齐胸襦裙。她循声转向朋友,眉梢轻抬,笑意刚刚出现。采用100–135mm远距观察、前景遮挡和低反差彩色负片质感。\n\n第二张「长安酒肆第七轮」:\n人物坐在胡姬酒肆的低案旁,刚放下鎏金耳杯便被同席朋友逗笑。乌发梳成乌蛮髻,穿孔雀蓝织锦内层、朱红半臂衣与深绯罗裙。采用35mm同席纪实视角,桌缘、胡瓶与宾客肩背自然进入前景。暖灯完整覆盖五官,背景人物分别投筹、交谈与听乐,不统一看镜头。\n\n第三张「乐坊舞毕后的彩色余息」:\n最后一支舞结束,人物在乐坊侧室罗帘旁停步调整呼吸。头发梳成唐代舞后高髻,一束发丝略微松散。暗红与孔雀绿舞衣采用不透明内层,披帛和袖缘保留轻盈层次。她循声望向帘外,嘴唇轻启,疲惫、满足与意外同时存在。采用50mm后台近景、柔雾彩色负片、暗红与灰蓝空间分层。\n\n第四张「马球越线后的回身」:\n盛唐女骑手下马后快步走向休息区,听见同伴呼喊便在步伐中回身。头发梳成左右清晰的双螺骑装髻,穿暗红内层、石青窄袖袍、深色分裁裙裤与皮革长靴。仅保留一只手握住从画面下缘进入的球杖杆身,另一只手不入镜。采用135mm长焦、约1/40秒彩色慢门追随,主体脸部清楚,马匹、旗幡与扬尘形成方向性色带。\n\n光线:\n四张均以宽面自然光或有罩室内灯连续照亮五官。环境色和灯笼只服务空间层次,不在脸部形成圆形光斑、碎光、花影或局部漂白。\n\n质感:\n真实彩色摄影,保留毛孔、细绒毛、自然血色、衣料重量和轻微胶片颗粒。借鉴经典彩色纪实、柔雾时装与运动摄影的画面组织方法,但不复制具体作品。拒绝黑白、近黑白和统一暖金商业滤镜。\n画面右下角加入低调小号手写签名「{argument name=\"签名\" default=\"Zhong\"}」。\n\n负面提示:\n拼图,多分镜,人物身份漂移,现代马尾,泛古风头冠,廉价影楼古装,统一摆拍表情,背景人物动作重复,复杂手势,双手承担多个动作,多余手臂,多余手指,错误手指,脸部光斑,彩色投影,塑料皮肤,黑白照片,现代场景,乱码文字,其他签名。",
"n": 1,
"size": "1024x1024",
"quality": "high",
"output_format": "png",
"background": "opaque",
"moderation": "auto"
}Performance
GPT Image 2 API performance and availability
Live Flatkey request telemetry appears here when enough image-generation traffic is available.
Activity
GPT Image 2 API usage and generation activity
This chart reflects live Flatkey generation requests and remains unreported until enough traffic is collected.
Pricing
GPT-image-2 Pricing
GPT-image-2 pricing here follows the current Flatkey catalog unit, not an assumed OpenAI direct tariff. Set image size, quality and output count before estimating cost. For edits, supply reference images and describe what must stay unchanged.
Add credits
Use the same Flatkey balance and API key across image, video, audio, and text models.
- Model Type
- Text to Image
- API
- /v1/images/generations
- Billing basis
- image
- Outputs
- 1–10
GPT Image 2 image generation and editing capabilities
Use the documented image workflows for new visuals, reference-led edits, and production-ready creative variants.
Text-to-image creation
Generate new product, editorial, and campaign visuals from a structured natural-language prompt.
Reference-based editing
Supply an image input to revise an existing visual while preserving important subject and composition details.
Flexible image composition
Create square, portrait, or landscape assets with flexible image sizes for different placements and channels.
High-fidelity visual input
Combine text and image context in one workflow for richer visual direction; audio and video inputs are not supported by this model.
Compare
GPT Image 2 and GPT Image 1: image API controls
GPT Image 1 values are not verified in this fact set, so only documented GPT Image 2 fields are shown.
| Capability | GPT Image 1 | GPT Image 2 |
|---|---|---|
| Endpoints | Unknown here | /v1/images/generations; /v1/images/edits |
| Sizes | Unknown here | 1024x1024, 1536x1024, 1024x1536, auto |
| Formats | Unknown here | PNG, JPEG, WebP |
| Quality/background/moderation | Unknown here | Documented above |
| Image quality ranking | Not asserted | Not asserted |
Choose the right request before you integrate
Review the supported workflows, live cost for the current setup, and the exact parameter contract used by Flatkey.
Calculated from the current Playground settings and the live catalog rate when a fixed unit price is available.
- Current setup
- 1024x1024 · high · 1×
- Live rate
- $0.0066 / image
- Estimated request
- $0.0066
The final charge follows the submitted request and your account group. Token-priced or tiered requests cannot be reduced to one fixed estimate.
Workflows and Model ID
Use only the workflows marked available. Flatkey keeps one public model ID while the request body selects the input workflow.
Text to Image
Generate a new image from a text prompt and output controls.
- Model ID
gpt-image-2- Endpoint
/v1/images/generations
Reference-guided Image
Use reference images for edits, variations, or identity and composition guidance.
- Model ID
gpt-image-2- Endpoint
/v1/images/generations
Parameter compatibility
These values come from the same model configuration used by the Playground and request preview.
nImages1–10
Default: 1sizeSize1024x1024 · 1536x1024 · 1024x1536 · auto
Default: 1024x1024qualityQualityauto · high · medium · low
Default: highoutput_formatOutput formatpng · jpeg · webp
Default: pngbackgroundBackgroundopaque · auto
Default: opaquemoderationModerationauto · low
Default: autoimageReference limit4 files
Endpoint: /v1/images/generationsFrom first test to production
Use the same request contract in the Playground and API, then handle validation, balance, task state, and output retrieval explicitly.
Request lifecycle
- 1
Validate the model, prompt, references, and output settings.
- 2
POST the image request with the documented fields.
- 3
Read the generated output or asynchronous task result returned by the route.
- 4
Store the generated file and request metadata in your own workflow.
Production checks
Reject unsupported sizes, durations, ratios, formats, or reference counts before submitting.
Handle invalid keys, insufficient balance, rate limits, and account-group pricing separately.
Surface the rejected prompt or asset clearly so the user can replace only the failing input.
Keep the task ID, use bounded retries, and never treat a timeout as a successful generation.
Prompt library
GPT Image 2 prompt examples for product and marketing visuals
Start with a concrete subject, composition, output size, and brand constraint; adjust the request fields separately.
Generated with Image 2
Scene: A 16:9 sci-fi game loadout screen in a blue-purple spaceship hangar. A female armored character stands centered with a glowing violet rifle; armor slots line the left, three weapon slots line the right, and the lower slot is highlighted. Keep the HUD hierarchy, pose, lighting, and equipment geometry fixed for an exact equipment-switch frame. No readable words, logos, invented stats, or watermark. Composition and rendering: Deliver one 16:9 game-interface keyframe. Separate the character, held equipment and HUD into clear depth layers; keep slot spacing and icon scale consistent, with the selected item visually dominant. Preserve the scene-specific game art style. Let material edges catch the existing light without letting interface glow obscure the silhouette. Keep all equipment fully inside the frame. Avoid: warped perspective, malformed anatomy, merged objects, inconsistent shadows, excessive sharpening and watermarks. Follow the scene-specific text and branding restrictions above.
Generated with Image 2
Scene: A rainy night soccer broadcast frame under bright stadium floodlights. A black-uniform player strikes the ball while two white-uniform defenders close in, with the goal and a blurred crowd behind. Keep the broadcast camera angle, blank blue score bars, spray, and motion blur consistent; leave overlays abstract and text-free. No real teams, athletes, logos, or watermark. Composition and rendering: Deliver one 16:9 broadcast still at the decisive instant. Use coherent perspective for the playing surface and equipment, readable separation between competitors, and a softer crowd behind the action. Keep the principal subject sharp, with directional blur limited to fast extremities and spray. Preserve the specified broadcast angle and reserve overlay areas clear of the action. Avoid: warped perspective, malformed anatomy, merged objects, inconsistent shadows, excessive sharpening and watermarks. Follow the scene-specific text and branding restrictions above.
Generated with Image 2
Scene: A closed matte-black square smartwatch with a black strap rests on a wet glossy black tabletop. Water droplets catch a cool blue rim light and a soft reflection sits beneath the watch. Use a low three-quarter product camera and preserve the case, strap, highlights, and empty dark background. No readable branding, extra products, hands, or watermark. Composition and rendering: Deliver one 16:9 commercial product still. Keep the complete product silhouette inside generous crop-safe margins. Resolve edges, seams and material transitions precisely; shape highlights to describe volume without clipping bright surfaces or crushing dark details. Align reflections with the object and light source. Retain the stated camera angle and existing negative space; add no decorative props. Avoid: warped perspective, malformed anatomy, merged objects, inconsistent shadows, excessive sharpening and watermarks. Follow the scene-specific text and branding restrictions above.
Generated with Image 2
Scene: On a rain-soaked observatory roof at sunrise, a lone figure in a long dark coat walks toward a large telescope beside an open dome. Wet stone reflects the warm doorway light; cloud-covered mountains sit beyond. Hold the screen direction and end on the telescope and figure in the same wide composition. No readable text, logos, or watermark. Composition and rendering: Deliver one 16:9 storyboard keyframe, not a collage or an action sequence. Separate foreground, character plane and background through the existing light and atmospheric depth. Keep the stated pose and eyelines readable, and retain environmental landmarks for continuity. Match the described photographic, illustrated or puppet medium; give fabric and surfaces texture appropriate to that medium. Avoid: warped perspective, malformed anatomy, merged objects, inconsistent shadows, excessive sharpening and watermarks. Follow the scene-specific text and branding restrictions above.
Generated with Image 2
Scene: In a bright home kitchen, a surprised cook in a blue shirt and cream apron reaches toward pancakes, a frying pan, flour, bowl, and whisk suspended midair. Freeze the comic cause-and-effect moment with believable weight, warm daylight, and a clear path for each object. No injury, logos, readable words, or watermark. Composition and rendering: Deliver one 16:9 physical-comedy still at the described instant. Make the cause of the gag readable through body balance, grip and object spacing. Separate the reaction from the airborne props; retain believable gravity and contact points. Keep the face and important props sharp, with only restrained directional blur on fast edges. Preserve the stated lighting and room or stage layout. Avoid: warped perspective, malformed anatomy, merged objects, inconsistent shadows, excessive sharpening and watermarks. Follow the scene-specific text and branding restrictions above.
Generated with Image 2
Scene: A rainy period harbor platform shelters people in era-appropriate coats and umbrellas as a vintage green tram arrives on wet tracks. Wooden waterfront buildings and misty mountains sit behind the reflections. Preserve the historical clothing, tram shape, rain direction, and stable wide composition; no readable signage, logos, or watermark. Composition and restoration: Deliver one 16:9 archival-style still. Preserve the original framing, period-specific construction and restrained tonal range; retain monochrome, sepia or faded color as described. Recover modest detail in faces, clothing and architecture without synthetic sharpening. Keep fine grain and authentic aging consistent across the frame; repair damage without replacing historical features. Avoid: warped perspective, malformed anatomy, merged objects, inconsistent shadows, excessive sharpening and watermarks. Follow the scene-specific text and branding restrictions above.
Why use Flatkey for GPT Image 2?
A repeatable GPT Image 2 workflow for product images
Keep image controls and image-input token dimensions together so a prompt test can become a reproducible API request.
All documented controls in one place
Set n, size, quality, output format, background, and moderation before you hand the request to the console.
Image input stays explicit
The catalog exposes token and image-input dimensions; this is not a separate fee for output pixel size, and the final amount depends on request and usage fields.
Marketing-ready starting points
Use product, ad, and storyboard examples as editable starting prompts rather than generic image filler.
No unsupported promises
The page does not promise transparent output, free generation, or a fixed per-image price when those facts are not verified.
API
GPT-image-2 API
Four ways in, all on the same key and the same model catalog. Pick one to see a runnable example.
curl https://router.flatkey.ai/v1/images/generations \ -H "Content-Type: application/json" \ -H "Authorization: Bearer sk-***" \ -d '{"model":"gpt-image-2","prompt":"Create a high-quality product image with GPT-image-2: clean composition, precise lighting, strong subject focus, and realistic detail.","n":1,"size":"1024x1024","quality":"high","output_format":"png","background":"opaque","moderation":"auto"}'
FAQ
GPT Image 2 API
pricing and prompt questions
Answers about GPT Image 2 API access, image-generation fields, token and image-input pricing, formats, and background behavior.





