News from the prompt

Your tools and pipelines now work inside the AI assistants you already use. Ask Claude to remove a background, or ChatGPT to run your pipeline on a handful of images, and it calls apiai.me for you, paid from your balance like any other run.

The new “Use in AI tools” page in your dashboard has the setup for each:

• Claude.ai and the ChatGPT desktop app: add apiai.me, sign in, click Allow. No API key to copy.
• Claude Code, Cursor, VS Code, Gemini CLI and Codex: one command or one click, with your key already filled in.

Every app you connect is listed on the same page, and you can disconnect it there at any time.

You can now sign up and sign in with your Google account. Press “Continue with Google” on the login page, pick your account, and you’re in. No code to wait for.

If you’re already signed into Google, the homepage offers a one-click “Use apiai.me with google.com” prompt in the corner.

Already have an account? Google signs you into it, even if you typed your Gmail address with different dots when you registered. Team invitations work the same way: continue with the Google account the invite was sent to and you join the team straight away. The emailed code still works exactly as before.

When you add someone to your team, the invitation they get now says who invited them, what apiai.me is, and the three steps to get in: open the link, ask for a code, type it in. There’s no password to set. It also mentions that the code can land in a junk folder and is readable right from the subject line, which was catching a few new teammates out.

Replies to the invitation go straight to you, the person who sent it, so a colleague who’s unsure can just hit reply and ask. And the code email for joining now names the team it’s for.

Every endpoint is now available at a versioned address — for example https://apiai.me/api/v1/process/your-tool-slug. That’s the path to use going forward.

Your existing calls to the unversioned /api/... keep working exactly as before — nothing to change today. Versioning just lets us keep improving the API safely: within v1 we only make backward-compatible additions, and anything that could affect existing integrations will ship under a future version with plenty of notice.

The Funds page now lists your past top-ups — date, amount, and status — each with a link to a downloadable receipt.

Handy for expense reports and for reconciling your billing at a glance, without digging through email.

Eval results now carry the ID of the API run that generated the output, shown in a new Run column. A score is no longer floating on its own — you can trace it straight back to the exact run in your dashboard and Recent Activity.

Runs started manually or from a monitor simply show no ID.

When you run a tool or pipeline, the result footer now shows a short run ID — click it to copy. The same ID is saved alongside each result in your history and shows up in the Recent Activity list on your Analytics page.

Handy when you want to point support at a specific run, or line a result up with your own records.

Two improvements when building pipelines:

• For optional image inputs (e.g. Kling 2.5’s end_image), you can now pick Skip or Default in the source dropdown instead of having to pass a value. Build pipelines around the inputs you actually use.

• Quality Gate now shows where its image input comes from — the previous step, your API upload, or an exposed parameter — so it’s clear at a glance that you can drop one in the middle of a flow, not just at the end.

The parameter list in custom pipelines now keeps the same order every time the page loads. Useful for any pipeline with multiple image inputs (e.g. mask + product photo) where rows used to drift between sessions — the slots stay where you expect them, and your uploads land where you intended.

You can now upload .heic files straight from your iPhone — no need to convert to JPEG first. The Format Converter tool also accepts and produces HEIC, with proper handling of HDR / Display P3 colour so results don’t come out crushed or muddy.

Large iPhone shots are auto-resized to a sensible web-friendly size during conversion, so a 12 MP photo doesn’t land as a 5 MB output unless you ask for that.

A few stubborn spots in light mode were rendering nearly-transparent or white-on-white — the prompt editor modal, parts of the Invite Users page, and a handful of sub-surfaces in the slideout. All fixed, the page reads cleanly in light mode now.

If you spot anything else that doesn’t look right when light mode is on, the contact form is open.

A few quick wins on the Batches screen:

• Your own pipelines now sit at the top of the workflow dropdown, then curated pipelines, then single-step tools. The thing you actually built and want to run gets prime real estate.
• True/false parameters render as proper toggles instead of “true” / “false” text inputs.
• Parameter rows match the layout you already know from the API tester: description on the left, control on the right, room to type.
• Need a prompt in your batch? Prompts now show up as a proper text area (and click into the same focused editor used in the API tester).

The upload limit is now 50 MB, up from 20 MB. Anything over about 20 MB gets gently resized server-side before it goes on to the model — typically landing around 1–2 MB at 2048 px on the long edge. Quality stays great for almost every workflow, and you get a heads-up header back telling you what we did so you can tighten your client if you want.

Same image, same SKU filename, smoother turnaround. Particularly handy if your catalog photos come straight off a phone.

If something is too big to handle (over 50 MB), you now get a clear, friendly message that tells you the size limits in plain MB instead of a wall of bytes.

Big batches are sturdier now. If a single file in a run is corrupt or in an odd format, that one item is marked failed and the rest of your batch sails through. No more one bad apple holding up the whole job.

Downloads of finished batch results are a little quicker too.

Upload summer-dress-01.jpg and the result comes back as summer-dress-01, not a random ID. When you’re running hundreds of items through a pipeline, matching outputs back to your SKUs or source files is now effortless.

This applies across single requests and batch jobs.

We reworked how pricing is presented so it’s easier to see what a run costs and what your credits actually get you, with a refreshed pricing page to match.

Eval profiles can now take typed input slots: define the inputs your evaluation needs once, write the rubric against them, and reuse the same profile across many runs instead of rebuilding it each time.

Each result records exactly which inputs it ran with, so your comparisons stay honest.

Pipelines can now take named parameter inputs, so the same flow runs with different values per request instead of hardcoding them. Build one pipeline, reuse it everywhere with the knobs that matter exposed.

Video models now accept a wider range of image inputs, so feeding in a starting frame or reference image is more flexible. We also tidied up billing for video runs along the way.

Google’s Veo video model is now available through apiai.me. Generate video from text or an image with the same one-call simplicity as everything else.

We tuned our video generation for noticeably higher quality results. Same call, better frames.

The whole site and dashboard now come in both light and dark. Choose whichever suits you and we’ll remember it next time you’re back.

We launched AI KNOW, practical writing on AI pipelines, image and video generation, and shipping production AI features faster. You’ll find it linked in the footer under Resources.

Batch jobs now accept uploads up to 200MB and chunk large sets automatically. Push a much larger product catalog through in a single job instead of splitting it up yourself.

Pipelines and flows can now fire an evaluation automatically when they finish, so you get a quality read on your outputs without wiring up anything extra. Especially handy for catching regressions across large runs.

Our Canva app is live. Generate images, run pipelines, and pull in video without leaving your Canva project. Bring AI output straight into the designs you’re already working on.

We now run a public MCP server, so you can drive apiai.me straight from MCP-compatible tools and AI agents. Point your client at it and your pipelines show up as callable tools, no glue code required.

Built a pipeline that works? Share it with teammates so everyone runs the same flow instead of rebuilding it from scratch. Comes with a broader cleanup of team and organisation handling, including clearer member management.

You can now call OpenAI’s image models through the same single endpoint as everything else. Drop them into pipelines alongside Replicate, Gemini and the rest, no extra setup or separate keys to wrangle.