Blog
AI Image & Video API Guides
Hands-on model tests, production tutorials, prompt techniques, and current API pricing for generative media teams.
Most Realistic AI Image Generator: 4 Models Tested Head-to-Head
Seedream 5.0 Pro produced the grittiest, most texture-dense skin of the four models and is our pick for the most photorealistic AI image generator in this test. Nano Banana 2 was a close second on realism with slightly smoother skin, and its 2K tier ($0.076) is actually cheaper than its 1K tier ($0.085) — an unusual, confirmed pricing quirk. Qwen Image 3.0 Pro is the budget realism pick at $0.0497 per 1K image, with visible pore-level detail that punches above its price. FLUX.2 landed in the middle on realism but prices by megapixel instead of resolution tier ($0.05/1MP), which makes it the hardest of the four to compare on a like-for-like basis. All four models were run through the same 85mm f/1.8 golden-hour portrait prompt via the HiAPI /v1/tasks endpoint, with zero prompt edits between models.
heygen-avatar-v Prompt Recipes: Copy-Paste Prompts With Real Outputs
Two real heygen-avatar-v clips embedded with the exact JSON payload used to render each one. Covers the required prompt/audio_url field, avatar and voice presets, aspect ratio, resolution, and caption behavior. Explains a real gotcha: filmed avatar presets ignore background color/image overrides entirely. Includes a full /v1/tasks create-poll-download walkthrough and a $0.15/second pricing table.
How to Use gpt-6-astra via the hiapi API: curl, Python, and a Working Request
gpt-6-astra is a reasoning text model live on hiapi, reachable only through POST /v1/responses. The model requires "stream": true — non-streaming requests always fail with invalid_request. Includes working curl and Python examples that parse the SSE event stream for the final answer. Covers auth errors (401 permission_denied), stateless requests via store:false, and prompt caching.
Veo 3.1 Lite Image-to-Video: Turn Photos Into Short-Form Video via the hiapi API
veo-3.1-lite/image-to-video turns one still photo into a native-audio vertical clip — ideal for TikTok and Reels workflows where you need to control composition first. The task API takes five fields: prompt, a single image_url, duration (4/6/8s), resolution (720p/1080p), and generate_audio — extra fields are rejected. A 4-second 720p clip with audio costs $0.40 ($0.10/s); dropping audio brings it to $0.24 ($0.06/s). In our test, the clip rendered in about 55 seconds from submission to a ready download link. Full working curl and Python examples are included, along with the real prompts used for both the still and the motion pass.
Recraft Remove Background API: A Working Example
recraft/remove-background takes one image URL and returns a transparent PNG cutout — no prompt or strength parameter needed. The schema is strict: image_url is the only accepted input field; extras like format or response_format get rejected with a 400. Output is a temporary URL (expireAt ~1 week) — download the bytes immediately instead of storing the temp.hiapi.ai link. Flat $0.01/image pricing, and two distinct error envelopes (schema errors vs auth errors) that batch code needs to check separately.
How to use 851-labs/background-remover via the hiapi API: curl, Python, and a working request
851-labs/background-remover runs on hiapi's unified task API: POST /v1/tasks with model and input.image_url, poll GET /v1/tasks/<id>, download the transparent PNG from data.output[0].url. input.image_url is the only field the model accepts — a public http(s) URL. There is no prompt, mask, or aspect-ratio option, and any extra field is rejected outright. Use callback.when="final" for batch pipelines instead of polling; key handlers on taskId since delivery is at-least-once.
Qwen Image 3.0 Image-to-Image: Turn One Product Photo Into a Full E-Commerce Set
qwen-image-3.0/image-to-image can turn one studio product photo into multiple lifestyle scenes without a reshoot. In our test, product shape, material, and color stayed consistent across three different scenes. A single reference image was enough for a simple hero object; the model supports up to three. Base tier costs about $0.036/image plus a small per-reference fee, versus a full reshoot per scene. The workflow scales with a simple loop: same reference photo, one new scene prompt per call.
How to Use wan3.0-video via the hiapi API: curl, Python, and a Working Request
Minimal, verified curl and Python examples for creating and polling a wan3.0-video task on hiapi's unified async API. Full input schema table (prompt, duration, aspect_ratio, resolution, audio, reference_image_urls, reference_video_urls) with three real 400 error responses. Clears up a common mistake: the correct image-to-video field is reference_image_urls, not image_urls. Production notes on callbacks, idempotency keys, output storage/expiry, and error handling.
Veo 3.1 Lite Text-to-Video Prompt Recipes: 2 Copy-Paste Prompts With Real Outputs
Two copy-paste text-to-video prompt recipes for veo-3.1-lite/text-to-video, each shown with the real, first-take clip it produced — no re-rolls. Covers the model's optional native-audio toggle: an ambient scene with layered sound, and a silent vertical product loop shot to run muted. Includes the validated task API request shape and real per-second pricing across all four resolution/audio combinations — the two clips here cost $0.64 total.
Veo 3.1 Lite Image-to-Video Prompt Recipes: Copy-Paste Prompts With Real Outputs
veo-3.1-lite/image-to-video animates a single still image into a 4-8 second clip, with optional native synchronized audio. Three recipes shown end-to-end: a portrait turn-and-smile, a product reveal spin, and an environment mist-lift camera push. generate_audio is a real cost switch — it moves 720p from $0.06/s to $0.10/s, and 1080p from $0.10/s to $0.14/s. The schema takes a single image_url (no multi-image blending) and a fixed duration enum of 4, 6, or 8 seconds. A working Python submit/poll/download snippet is included so you can reproduce any recipe against the live API.
Build with the models you just read about.
Compare image, video, and audio models, then use the same API workflow in production.