API Tango Papamuri Angitu mō Ngā Kaiwhakawhanake
Whakaurutia te tango papamuri kaute powered AI ki āu tono mā tō mātou API tūhonohono me te waiawaia.
Whiwhi Kī APITāuta Ngāwari
Whakauruhia te tango papamuri ki roto i tō taupānga mā te tuhinga i te hārakirangi kape. Ko te mātuhanga tautuhi API me ngā SDK mō ngā rākau ngākaukahua ka whakaēhua i te taapiri.
Whakapunga Whāwhā mō Ngā Tono Rerekē
Whakaorahia te tukurua o te tukanga tango papamuri ki ō hiahia. Whakaritehia ngā taurangi, runga atu i te tomokanga ki ngā whakatakotoranga rerekē, te whakakapi papamuri tūturu hoki.
Ngaiotanga Mōtika Mahingi
Kua hangaia mō te whakawhiti me te tere. Ka taea e tō mātou API te whakapai i ngā patai puna poraka ā ia rā me te rahinga pokare, mēnā kei te pai tonu tō taupānga ahakoa ana ngā haurangi pae taumaha.
Hohiora Āhuritanga Pūmanawa i ō Tono
Whakamanahia ngā kiritaki me ngā kaha whakatairanga whakapenga. Mai i ngā tūao e-tauhokohoko, ngā pānga pāpori tū or ngā tū manamana, kāore he rohe mō ngā āheiratanga ki te tango papamuri API.
Ngā Utauta e Tūtohu Ana mō ngā Kaiwhakawhānui
How a small dev rōpū shipped a profile-whakaahua cropper feature in one sprint
A four-person development rōpū building a hobby-mākete app needed a profile-whakaahua feature that turned a user's casual phone shot into a mā catalog-grade avatar. The PM wanted it in the panuku sprint, the kaihoahoa wanted on-waitohu backdrops the user could pick from, and the pūnaha rōpū wanted no new server bills. A traditional integration would have meant a paid API key, a new microservice, and a queue.
The rōpū wired the ētita's in-browser cutout into the existing tukuna flow as a client-side step. The user picks a whakaahua, the cutout runs locally on their device, the user picks one of three waitohu-aligned backdrops, and the resulting JPEG goes straight to the same R2 bucket the rest of the tukuna flow uses. No server-side tukatuka, no key rotation, no per-request billing. The whole feature shipped in 480 lines of code, including the picker UI and the analytics events.
The feature went live at the end of the sprint, processed 14,000 avatars in the first marama with no extra infrastructure cost, and dropped the rōpū's profile-completion rate from 31 percent to 58 percent because the picker felt like a curated experience instead of an awkward tukuna field. The pūnaha bill stayed flat. The rōpū kept the same pattern in mind for a future hua-rārangi whakaahua step.
"We needed an avatar cropper that didn't add a server-side service or a paid API. Wiring the in-browser cutout into our tukuna flow took one sprint and shipped at zero marginal cost per user. The pūnaha rōpū noticed our request graph didn't change."
"I'm the only engineer and I needed a profile-whakaahua step that didn't pull in a third-party SDK we'd have to babysit forever. A client-side cutout meant I shipped the feature, then forgot mō it. No keys to rotate, no rate limits, no tautoko tickets six marama later."
"Bundling a heavyweight SDK into a starter template makes the whole project feel bloated. The browser-side approach means contributors can fork the template and not need to set up a third-party account. Adoption of the whakaahua step is up since I switched."
Picks that fit a kaiwhakawhanake workflow
Common questions for kaiwhakawhanake
Is there a stable API for the in-browser cutout, or do I need to embed the ētita iframe?
The ētita exposes a small JavaScript surface that you can call from your own page once the model is loaded. The cutout returns a Blob you own, so you can pipe it directly to your existing tukuna pipeline. The model loader handles caching across sessions via the Cache API, so the hēkona visit is tere. There is no iframe required and no postMessage handshake, the function is invokable like any other client-side whakaahua operation.
What's the cold-start cost of the model on a first-time visitor?
First-load fetches the WASM runtime and the model weights, which together are roughly 30 MB on the wire. A hou broadband connection gets that in two or three hēkona; a slow mobile network closer to ten. Subsequent visits hit the Cache API and start instantly. For latency-sensitive apps, a preload hint in the HTML head warms the cache before the user reaches the whakaahua step. Server-assisted fallback is available for devices that can't run the model locally.
Are there usage limits or quotas if I integrate this into a commercial hua?
The browser-side pipeline runs on the user's device, so there is no per-request quota and no rate limit to negotiate. Server-assisted fallback for the rare device that cannot run the model locally has its own quota documented separately. For high-volume commercial integrations the recommendation is to handle the local-cutout path as the default and surface server fallback only on capability detection failure, which keeps cost predictable as you scale.
Ship a whakaahua feature without adding a service
Wire the in-browser cutout into your existing tukuna component, keep the kōnae on the user's device, and pipe the hua straight to your storage.