Blog
AI Image & Video API Guides
Hands-on model tests, production tutorials, prompt techniques, and current API pricing for generative media teams.
GPT Image 2.5 vs Nano Banana 2: 9 Real Outputs
Both Image 2.5 models followed the two-stone requirement; Nano Banana 2 added a third stone. All three rendered the main Chinese poster copy correctly, but generated orange slices instead of peel. All three changed the shared image background; Nano Banana 2 changed the bottle color more noticeably.
HiAPI Editorial
Comparison · Sep 9, 2026 · 10
Using gpt-6-astra for e-commerce product images via the hiapi API
GPT-6 Astra is a text-only reasoning model — it can't generate the product photos, but it can generate everything written around them. Use JSON schema output on the Responses API to turn spec sheets into structured, ready-to-publish listing copy at scale. Build a store=false support assistant by replaying conversation history yourself, with function calls for order lookups. Standard pricing is $2.50/1M input and $12.50/1M output tokens; crossing 272,000 input tokens per request switches the whole call to the long-context tier. Pair this with an image model like GPT Image 2 for the visual side of the same product listing.
GPT Image 2.5 Flare vs Sunburst: Which to Use?
Start with Flare for exploration; try Sunburst for edits with strict preservation requirements. Both models generate and edit; compare them at matching quality settings. Choose using actual outputs, cost per accepted image, and elapsed time.
GPT Image 2.5 API: Generate, Edit, and Migrate
Flare and Sunburst support generation and editing through the async task API. Migration requires checking reference fields, dimension enums, and old route settings. Wait for a terminal state, save outputs, and verify costs against current settings and billing.
Recraft Background Removal for E-Commerce Product Images
recraft/remove-background takes a single image_url and returns a transparent-background PNG cutout, with no prompt or mask to configure. A real batch run against the live hiapi API produced a clean product cutout, shown alongside the original photo in this article. The model costs $0.01 per image as of 2026-09, so a 500-SKU catalog with one shot each costs about $5 to fully cut out. A reusable Python script fans a list of product photo URLs across a thread pool and returns the cutout URL for each source image. The narrow, prompt-free input makes it practical to run unattended across large product catalogs rather than one image at a time.
GPT Image 2 vs 2.5: Image Quality, Editing, and When to Upgrade
Image 2.5 focuses on reference fidelity, targeted edits, and consistency across revisions. Flare suits fast creative work; Sunburst prioritizes precise editing. Choose by task and keep Image 2 for workflows that already produce reliable results.
How to Use deepseek-v4.1-flash via the hiapi API: curl, Python, and a Working Request
deepseek-v4.1-flash is a Preview reasoning model on hiapi; the public model ID stays stable even though responses show a rotating pinned-build name. Minimal working curl and Python examples for POST /v1/responses, including the exact request/response shape. reasoning.effort for this model is none | high | max — not the low/medium/high scale used elsewhere. Requests are stateless: replay the full input array every call, no previous_response_id support. Streaming uses semantic events (response.output_text.delta ... response.completed), not a Chat-Completions-style [DONE] marker.
Turn a Script into a Talking-Head Short with heygen-avatar-v and hiapi's API
heygen-avatar-v drives a preset HeyGen avatar from a text script or audio file, returning a lip-synced clip with continuous idle motion (not a static talking photo) plus an optional synced caption track. This piece walks the full /v1/tasks pipeline for short-form output: a real generated clip, two gotchas (caption is an .srt sidecar not hardsubs; background only works on illustrated avatars), and live per-second pricing. Cost is metered at $0.15 per second of output video with no duration parameter, so script/audio length — not a config value — determines your bill.
How to Use deepseek-v4-flash-vision-exp via the hiapi API: curl, Python, and a Working Request
deepseek-v4-flash-vision-exp is a reasoning-capable text-and-vision model on hiapi, reachable via the standard POST /v1/chat/completions endpoint. It reasons before answering, so message.content can come back null if max_tokens is too low to cover both the reasoning pass and the final answer. It supports vision input (image_url), JSON mode (response_format), OpenAI-style tool/function calling, and both streaming and non-streaming responses.
How to Increase Image Resolution with hiapi's API: a 4K Upscaling Workflow
hiapi has no dedicated pixel-preserving upscaler, but image-to-image models like seedream-5.0-lite can redraw an existing image at 2K/4K resolution. This recipe gives a complete, runnable Python (and curl) example: submit a task, poll or use a callback, and download the higher-resolution result. Covers model choice (seedream-5.0-lite, gpt-image-2 @ext, wan2.7-image), idempotency, and the two distinct error shapes (400 vs 401).
Build with the models you just read about.
Compare image, video, and audio models, then use the same API workflow in production.