Blog
AI Image & Video API Guides
Hands-on model tests, production tutorials, prompt techniques, and current API pricing for generative media teams.
Best AI Image Generation APIs in 2026
There is no single best image API — the right model depends on the job; one integration should give you all of them. hiapi's image roster spans 70× in price, from flux-schnell ($0.005) to gpt-image-2-pro ($0.35). Same-prompt test: gpt-image-2 and Nano-Banana-2 render on-image text cleanly; FLUX 1.1 Pro is weaker on dense text. gpt-image-2 ($0.03) is the best price-to-instruction-following default.
qwen-image-2.0 prompt recipes: six copy-paste prompts with real outputs
qwen-image-2.0 is one of the few text-to-image models that renders multi-character Chinese and short English headlines cleanly inside the picture. It's on hiapi at $0.025 per image, flat, with a 2K default output and the full set of common aspect ratios from 1:1 to 21:9. Six tested recipes are included — bilingual storefront, ink landscape with calligraphy, modern poster, Pixar character, editorial illustration, photoreal flat-lay — each with the exact prompt and hiapi input. All seven images in the article (one cover + six recipes) cost a total of $0.175 to produce.
ChatGPT Images 2.0 vs gpt-image-2: Which One Should Developers Use?
ChatGPT Images 2.0 (consumer) and gpt-image-2 (developer API) draw from the same model family but solve different problems. OpenAI's direct API is token-billed, so the same 1024×1024 canvas can run ~35× more expensive at high vs low quality. hiapi resells gpt-image-2 at a flat $0.03/call at 1K, with gpt-image-2-pro at $0.35 for the polished hero-shot tier. Three decision rules: explore in ChatGPT, ship in the API, escalate to pro only where polish is the product.
Best AI Video Generation API in 2026: Comparing hiapi's Four Video Models
hiapi ships four video generation models spanning almost 5× in per-second cost at 1080p: wan2.7-t2v / wan2.7-i2v at $0.167/sec, happyhorse-1-0 at $0.288/sec, and seedance-2-0 at $0.823/sec. wan2.7 (t2v and i2v) is the price floor for HD output on the platform — use it as your default unless your project specifically needs cinematic stylization or premium polish. seedance-2-0 is the only model that handles both text-to-video and image-to-video in one integration, plus offers a cheap 480p tier ($0.15/sec) that's great for storyboarding before HD render. Once you factor in real acceptance rates (1 in 4 takes kept), the cost gap between cheap and premium widens fast — making model choice a real unit-economics decision, not just an aesthetics one.
GPT Image 2 vs FLUX 1.1 Pro: Quality, Prompt Adherence, and Speed
GPT Image 2 wins instruction-following, text accuracy, multi-element compositions, and per-image price ($0.03 vs $0.05). FLUX 1.1 Pro wins raw speed by roughly 9.2× (~6.4s vs ~59s per image in our test) and photoreal portraits with dramatic lighting. FLUX 1.1 Pro made a spelling error in the text-rendering test (EYERY for EVERY). GPT Image 2 rendered all text correctly across the same prompt. On a complex eight-object flat-lay, GPT Image 2 placed every item with the right count; FLUX 1.1 Pro produced a stylish but inaccurate version. Numbers are from hiapi production endpoints as of 2026-05; sample size is small by design — six paired prompts, single image per model.
FLUX-Schnell Prompt Recipes: Copy-Paste Templates With Real Outputs
FLUX-Schnell on hiapi is about $0.005 per image and ~30 seconds end-to-end via the unified /v1/tasks endpoint — the cheap-and-fast tier of the FLUX family. Seven prompt recipes covering hero imagery, e-commerce flat lay, avatars, landscapes, isometric SaaS, food, and concept sketches — three with real outputs you can inspect. Includes the structural prompt pattern that consistently produces clean outputs on a 1-4 step model, plus the two API quirks (no resolution field; only five supported aspect ratios) that trip up first-time users.
Text-to-Video vs Image-to-Video API Workflow
Use text-to-video for storyboards, ideation, and prompt-first creative exploration. Use image-to-video when the subject, product, character, or first frame must stay close to a source asset. Seedance 2.0 is the broad default; Wan 2.7 Image-to-Video is the focused first-frame workflow. Video jobs should use callback-based completion in production.
How to Call Multiple Image Models with One API Key
Use one server-side API key for model testing, billing visibility, and production routing. Route model choice through the request body instead of hard-coding separate provider clients. Start prompt iteration on a lower-cost model, then move final assets to a higher-fidelity image model. Use callbacks or task polling depending on your production volume.
The AI Image Prompt Library: 10 Production-Ready Prompts by Use Case
Ten prompts organized by the five use cases that account for most production image work: e-commerce hero shots, social covers, avatars, product display, and editorial illustration. Every prompt is paired with the hiapi model and aspect ratio it was tested at, plus a one-paragraph explanation of why the wording works — so the prompts can be adapted rather than just copied. Pick the model by job: qwen-image-2.0 ($0.025) for volume stylized work, gpt-image-2 ($0.03) for text-heavy layouts, Nano-Banana ($0.05) for portrait consistency, flux-1.1-pro ($0.05) for photorealism.
Adding AI Image Generation to a Next.js App With the hiapi API
Image generation takes 60–120 seconds — too long for a single client request, so the app must split the work across a route handler that creates a task and a polling endpoint that surfaces progress without holding a serverless invocation. Calling hiapi directly from a React component leaks the API key to the browser. A Next.js route handler keeps the key on the server and gives you one auditable place to rate-limit and validate prompts. Output URLs returned by hiapi are signed and short-lived. Apps that store them directly will serve 404s on refresh — mirror every successful result to R2 (or S3 / Supabase Storage) before persisting. Cumulative Layout Shift on AI-image apps is fixed by reserving the result container's aspect ratio with CSS and matching it to the ratio sent to hiapi.
Build with the models you just read about.
Compare image, video, and audio models, then use the same API workflow in production.