Blog
AI Image & Video API Guides
Hands-on model tests, production tutorials, prompt techniques, and current API pricing for generative media teams.
Consistent Characters with Nano-Banana on hiapi: A Working Workflow for Storyboards and E-Commerce
Three Nano-Banana models on hiapi all tagged for character consistency: Nano-Banana ($0.05 flat), Nano-Banana-2 ($0.085 / $0.076 / $0.114 across 1K/2K/4K — 2K is cheaper than 1K), and Nano-Banana-Pro ($0.17 at 1K/2K, $0.30 at 4K). The technique is a character bible — a dense, repeatable, deterministic prompt prefix that anchors face geometry, skin/hair markers, wardrobe, and vibe across all renders. No reference image, no fine-tune, no LoRA. Storyboard demo: one engineer character ('Mira') across cover, model sheet, warm-tungsten workshop, and cold-overcast rooftop. Same beanie, same toolbelt, same beauty mark stays on the right cheek through both scenes. E-commerce demo: one fashion model ('Aria') in three full-body look-book shots — frontal hero, three-quarter angle, then a wardrobe swap to a camel double-breasted overcoat. Face, hair parting, beauty mark, and gold studs all hold across renders. Flat Nano-Banana rejects the resolution field with a 400 — only set it for Pro and -2. Output URLs have an expireAt; download bytes immediately rather than hot-linking. Total cost for all 8 images in the article: $0.52 (7 × $0.05 Nano-Banana + 1 × $0.17 Nano-Banana-Pro hero).
hiapi Image and Video Generation APIs: The Complete Platform Guide
hiapi is an image and video generation API gateway: one Bearer-auth endpoint fronts FLUX 1.1 Pro, Nano-Banana, GPT-Image-2, Qwen-Image, Wan2.7, Seedance, and HappyHorse. Pricing is flat per task — $0.02 for gpt-image-2-beta, $0.05 for Nano-Banana or flux-1.1-pro, up to $0.823 for a 1080p Seedance clip. No token billing, no surge. All generation runs through the async POST /v1/tasks endpoint. Same shape for text-to-image, image-to-image, text-to-video, image-to-video. Most image models accept input.prompt + input.aspect_ratio. qwen-image-2.0 uses a DashScope-style nested schema with literal pixel sizes — keep a model→payload adapter map. Four sample images in this post — cover illustration, photoreal portrait, CJK calligraphy signboard, layout poster — were all rendered through hiapi for a total of $0.155.
FLUX 1.1 Pro on hiapi: Capabilities, Pricing, and a Working Code Example
FLUX 1.1 Pro is on hiapi at $0.05 per image, flat across aspect ratios — no resolution multiplier. It is served on the async /v1/tasks endpoint, not /v1/chat/completions; use `aspect_ratio` not `size`. Photoreal portraits, product flat-lays, and short headline text render reliably; unprompted text on objects is still hallucinated. Older FLUX.1 Dev and the original FLUX Pro are not on hiapi — the catalogue carries only the 1.1 Pro flagship.
Inside Codex's Agent Loop: Lessons from Datadog and Sora's 28-Day Android Launch
The Codex agent loop is a simple control flow — prepare prompt, model inference, optional tool call, append result, repeat — that terminates when the model emits a normal assistant message. Three implementation details make it production-grade: prefix caching on the Responses API, tool-call results that live inside the conversation, and termination detected from output shape (not a turn counter). Datadog uses the loop for system-level code review on every PR across 1,000+ engineers; its incident-replay harness showed Codex catching ~22% of historical incident-related issues human reviewers had missed. OpenAI shipped Sora for Android in 28 days with four engineers, ~85% Codex-generated code, and a 99.9% crash-free rate — by scaffolding with exemplars and AGENTS.md, not by prompting 'go build it.' The same agent-loop architecture transfers to media-generation agents built on hiapi: stable system prompt + tool-call accumulation + critique-based termination, wrapping image and video endpoints.
GPT Image 2 Multi-Turn Editing and Style Consistency: A hiapi Capabilities Tour
gpt-image-2 (text-to-image) and gpt-image-2-image-to-image are two halves of the same workflow — first draft and surgical editor — both priced at $0.03 per call at 1K on hiapi. The two patterns that move the needle: a verbatim character bible reused across scenes for series consistency, and a first-render-then-edit two-call flow for any job that must look like an extension of an existing brand asset. Both variants speak the same async task API — POST /v1/tasks, then poll the task ID — the only difference is that the editing variant adds an input_urls array of reference images.
Does GPT Image 2 Really Nail Text? We Stress-Tested Signage, Posters and Labels
Across 24 text-heavy generations — including English headlines, CJK calligraphy, multi-string infographics, embedded brand blocks, currency, and dense UI mockups — every primary string rendered correctly. GPT Image 2 will sometimes add contextually appropriate text you didn't ask for (brand monograms, founding-year badges, scent names). Useful most of the time, undesired occasionally. The remaining ~1% error rate at production scale shows up on the smallest text — sub-12pt captions, dense legal copy, ingredient lists. Always proof at 100% zoom before shipping. If text matters, lock it in straight quotes. Smart quotes get rendered as smart quotes. Special characters (em-dashes, ampersands) work; single-character substitutions are the most common failure mode when they occur.
GPT Image 2 E-commerce Workflow: From Product Spec to Listing Set
Write one fixed product specification, then vary only the shot instructions to build a more consistent listing set. The workflow covers a white-background packshot, alternate angle, detail, lifestyle scene, and packaging with verified on-pack text. Three retained outputs include their exact prompts, aspect ratios, task IDs, and recorded task costs. A reusable Python script submits a complete shot list without repeating generic task-polling documentation.
GPT Image 2 Review: A Hands-On Honest Take
Text rendering is the real upgrade — English headlines, CJK calligraphy, in-image labels and prices all render correctly the first time. Photorealism is genuinely strong for product photography and detailed interiors. Portraits are good but occasionally read as slightly 'too symmetric'. Speed is a step backward from GPT Image 1.5: ~107 seconds per image versus 18–36 seconds. Plan workflows around it. Hard limit: no transparent backgrounds. If you need PNG cutouts, you're still going to need a second model or a background remover.
15 GPT Image 2 Prompt Templates With Real Outputs
Fifteen copy-ready prompts are paired with the real outputs retained from HiAPI's first-party generation tests. The collection covers product photography, product ads, posters, infographics, brand concepts, icons, food, and editorial illustration. Three product-ad recipes retain their aspect ratios, task IDs, filled variables, and recorded task costs. Each template separates reusable creative direction from the product, headline, scene, or subject you need to replace.
GPT Image 2 vs Nano Banana 2: Side-by-Side With the Same Prompts
GPT Image 2's marquee text-rendering advantage has narrowed — Nano Banana 2 renders English and CJK headlines just as accurately, and even adds contextually appropriate label text on its own. Nano Banana 2 produces more believable photorealism (portraits, product photography), while GPT Image 2 keeps a slight edge on editorial typography craft. Nano Banana 2 effectively ignores precise aspect-ratio requests — square or near-square is what you get. GPT Image 2 honors 3:2, 2:3, and 9:16 reliably. Price: GPT Image 2 is $0.03 per image, Nano Banana 2 is $0.085 — 2.8× the cost. Speed is comparable on average.
Build with the models you just read about.
Compare image, video, and audio models, then use the same API workflow in production.