Make ads by chat in Claude or ChatGPT: which connectors exist and what they cost

Here is the business problem. You make ads, or you make them for clients. Every new tool means another login, another menu, another thing to teach a colleague. A chat window is the one place everyone already knows how to use. So the idea of typing "make me a 10 second ad for this product" and getting a video back is attractive.

It is possible today, but not the way many guides make it sound. The chat assistant is the driver, not the engine. In this page I cover the cheapest way to try the idea, the realistic options (including ones that are not Higgsfield), what the chat route costs, and a short test with a pass or fail rule. I keep the connection steps brief and send you to the full Higgsfield connection guide for the click by click version.

This page may contain affiliate links from which Blur Studio earns commissions.

The short answer: Claude does not make video, a connected service does

Claude (made by Anthropic) and ChatGPT (made by OpenAI) are chat assistants. They write, plan and reason. For video and images they hand the job to another service. The hand-off works through MCP, short for Model Context Protocol, a standard way for an assistant to call outside tools. Think of it as a plug that lets the chat window press buttons on another site for you.

Three things follow from that.

• You still pay the video service. The chat assistant does not include video credits. You need a plan with the service that makes the clip.
• The assistant is only as careful with money as you tell it to be. Each service decides what its connector can do and whether it has a spending limit.
• The video is made by the service's models, not by Claude. Quality, rights and limits are the service's, not the assistant's.

If you only wanted a better script, a shot list or ad copy, you do not need a connector at all. Claude or ChatGPT on its own does that job for the price of your chat plan.

The cheapest way to try this before paying for anything

Start with the option that costs nothing and teaches you the most.

1. Use the service's own website first. Make one clip on the video tool's site, by hand. You learn what the tool can do and what a clip costs before you add a chat layer.
2. Write the ad brief in a plain chat with no connector. Ask Claude or ChatGPT for the product, audience, one claim you can prove, three shots and a voiceover line. Paste the result into the tool's website. This is already most of the value of "ads by chat".
3. Test the connector pattern on a free tier. Anthropic's help page says Claude Free allows one custom connector, so you can try the chat pattern with one service without upgrading your chat plan. Whether the video service itself has a free tier is a separate question, and Higgsfield does not (see below).

If a clip from the website does not meet your quality bar, the chat route will not fix that. It only changes how you ask.

Five ways to get a video from a chat window

These are the connectors listed on vendor pages on 3 November 2026. Prices change, so confirm on the live page before you buy.

Option	What it is	Chat route and cost (as of November 2026)	Best when
Website only	Use the video tool's own site, no chat.	Plan price only. No connector risk.	You make a few clips a month.
Runway	An AI video maker built for filmmakers and marketers.	Free: 125 one time credits. Standard $15 a month ($12 billed yearly). Pro $35 a month ($28 billed yearly), which Runway lists with its MCP. Max $95 a month ($76 yearly), also listed with MCP.	You already pay for Runway Pro or Max.
HeyGen	A tool that makes videos with an AI presenter (an avatar) who speaks your script.	HeyGen publishes a connector and stores videos in HeyGen Projects. Its plan prices were not checked for this page, so read HeyGen's pricing page.	You need a talking presenter, not scenery.
Higgsfield	An AI image and video platform that puts many models behind one account.	Needs an active paid plan. Every chat generation uses credits. See below.	You want many models and ad tools behind one login.
Other connectors	Artlist (stock footage and AI tools), OpenArt, fal (a model marketplace for developers) and similar.	Artlist lists several of these in its own comparison article. That is a rival's list, so treat prices and free credits as unconfirmed until you read the vendor page. For example, Artlist's table says Higgsfield offers free renders to start and lists Sora 2, while Higgsfield's own pages show a Free plan with no credits and an active paid subscription required for the connector, and Sora is not in its current model list. Take the rival's table as dated.	You want to try the pattern at zero cost.

Two honest points. First, if you already pay for Runway Pro or Max, Runway lists its connector with those plans, so adding a second subscription just to chat with it makes little sense. Second, the comparison in Higgsfield versus Runway applies here too: the video model that suits your product matters more than the chat window.

Where Higgsfield fits: what the connector does and what it costs

Higgsfield's help center says you can connect it to Claude, ChatGPT and other agents, and then ask for images and video in the chat. The facts that matter for a business owner, taken from Higgsfield's own help center and blog:

• A paid plan is required. The help center says an active paid subscription is needed. Higgsfield's Free plan has no credits. Be careful with guides that promise free starter credits through the connector. The official page does not say that.
• No API key. You sign in with your Higgsfield account.
• Every chat generation uses credits at standard rates. A credit is the unit Higgsfield charges in. Unlimited model access and free generations apply only on the Higgsfield website, not through the connector. If you bought a plan for the unlimited image models, chat requests will still draw down your credits. This is the most common surprise.
• No built in spending cap. Telling the assistant "do not spend more than 300 credits" is a request, not a lock.
• The assistant cannot read pictures you attach to the chat. For reference photos, the assistant opens an upload window, or uses an image link or an earlier generation.
• Results land in your account. They show in the chat and in your Assets tab, tagged as coming from MCP, usually within about a minute of finishing.
• Audio and website building are not available in ChatGPT. If you need voiceover through chat, Claude is the better host.

Higgsfield's plans as of 3 November 2026 are Starter at $19 a month with 270 credits, Plus at $59 a month ($47 billed yearly) with 1,200 credits, and Ultra at $129 a month ($99 yearly) with 3,000 credits. Check the live pricing page before you rely on these. My detailed breakdown is in Higgsfield pricing explained.

A cost example you can check

Higgsfield's own blog gives one example. A 10 second Seedance 2.5 clip (Seedance is a video model from ByteDance, available on Higgsfield) showed 70 credits at 720p and 160 credits at 1080p when the assistant quoted the price. These are vendor examples from one session, so your numbers may differ.

Here is how I would use them. This is illustrative arithmetic, not a record of what anyone spent.

Step	Credits (vendor example)
Six 720p drafts at 70 each	420
One 1080p final at 160	160
Total for one ad	580

On a plan with 1,200 credits a month, that is about two finished ads and a little left over, before any re-rolls. Higgsfield's help center also says Seedance 2.5 is not included in Starter, so a Starter plan may not run this example at all. Confirm which models your plan includes before you plan around one.

The habit that saves money is draft low, final high. Ask for 720p first and approve one higher quality version at the end. Ask for the price before any upscale, because Higgsfield's blog says upscaling is the exception that does not show its cost in advance. For more ways to stretch credits, read how to spend Higgsfield credits wisely.

Connect it in about ten minutes

The short version for Claude, as given in Higgsfield's help center:

1. In Claude, open Settings, then Connectors, then Add custom connector.
2. Name it Higgsfield and enter the address https://mcp.higgsfield.ai/mcp.
3. Click Connect and sign in with your Higgsfield account.

On a Claude Team plan, only Owners and Primary Owners can add a connector. Members then connect to it one by one. For ChatGPT, Claude Code and the command line tool, and for what to do when authorization fails or the connector is not being called, use the steps in the Higgsfield MCP, CLI and Claude plugin guide. In Claude Code (Anthropic's coding assistant), the agent can install the Higgsfield command line tool instead; that route suits people who already work in a terminal, and the guide covers it. If you prefer a built in agent with no outside chat, see Supercomputer and Canvas.

Keep spending under control when there is no hard cap

Because the connector has no cap, you put the brakes in yourself. Paste this at the start of every session:

Before you generate anything, tell me the credit cost and wait for my yes. Do not spend more than 300 credits in this chat. After each run, tell me how many credits are left in this budget.

Then add three habits.

• Do not click Allow always on the generate tool. Anthropic advises reviewing each tool approval and using Allow always only for servers you trust. A per run approval is your real spending gate.
• Check your Usage page after the session. Compare it with what the chat said it spent.
• Keep the account owner as the only person who connects it. A client or freelancer should get files from the Assets tab, not a shared login. If several people generate, look at the Team plan with separate seats rather than sharing one account. Higgsfield's pages do not say that sharing one login is allowed, so I would not do it.

A brief to paste for a first ad

A vague request wastes credits. A clear brief gets a usable draft sooner. Copy this and fill in the brackets.

Product: [name and what it does in one sentence]. Audience: [who]. One claim I can prove: [claim and where the proof is]. Length: [10 seconds]. Format: [9:16 for Stories or Reels]. Three shots: [1, 2, 3]. Voiceover line: [text]. Do not show any brand names or logos that are not mine. Quote the credit cost first and wait for my approval.

If you want many versions of one ad, the chat is not the best place to plan them. Read how to test 20 ad variations and how to test ad hooks first, then use the chat to make the clips.

A 30 minute test with a pass or fail rule

This is a plan you can run before you commit to any plan.

1. Pick one product and write the brief above.
2. Set a budget, for example 300 credits, and paste the spending rule.
3. Ask for three different 720p drafts, approving each cost before it runs.
4. After each run, check that the clip appears in your Assets tab.
5. Compare the Usage page with what the chat reported.

Pass if all of these are true: the cost is shown before every run, the clips appear in Assets within about a minute of finishing, the Usage page matches the chat, and at least one of three clips is usable without a re-roll.

Fail if the cost appears only after the run, credits drop by more than the amount shown, or none of the three drafts is usable. If it fails, go back to the website route and use the chat only for scripts.

Check the output before it goes to a client

• Does the product look like your product? AI video often changes shapes, labels and text.
• Is every claim in the voiceover or on screen true and provable?
• Does any face or voice resemble a real person? You need consent from any real person whose likeness or voice is used. Check local law and the platform's current rules.
• Does any logo or brand name appear that you do not own? Higgsfield blocks many of these, but do not rely on the filter.
• Does the platform where the ad runs require an AI label? Check its current rules. An AI tool does not remove the need to disclose where a rule requires it.
• Should this moment be real footage? If the ad claims a real result, a real customer or a real before and after, film it. Keep real footage where a claim must be real.

Higgsfield's terms say outputs are not guaranteed to be unique, so never promise a client that nobody else will get the same look. Free account outputs carry a watermark, and the pricing table marks commercial use as not included for Free while the help center says use is not restricted to paid plans. For client work, use a paid plan.

Is it safe to add a custom connector?

A connector gives a chat assistant a way to act on your account, so treat it like giving a helper your login. Anthropic warns that a malicious connector could try to trick the assistant with hidden instructions, which is called prompt injection. Only add connectors from companies you have already decided to trust, read each approval request, and remove connectors you stop using. Higgsfield may use content to train and improve its models except under Enterprise agreements, so do not paste a client's confidential brief into any AI tool before reading its data terms.

Who should skip the chat route

• You make fewer than a handful of clips a month. The website is simpler and has no spending surprise.
• You need frame level control or a precise multi shot edit. Use a normal video editor.
• Your client approves every file in a portal. Chat history is not an approval trail. See a client approval workflow.
• You want a presenter talking to camera. HeyGen's connector is built for that, and the agency tool guide covers other options.
• You are a developer who wants key based billing and the biggest model list. A model marketplace such as fal fits better. To build it into your own tool, see API options and costs.

How I'd approach this

I would make one clip on the website first, then repeat it by chat with a 300 credit budget and the test above. If the chat route passes, I would keep it for drafts and keep finals in a normal editor. If you choose to try Higgsfield, you can start through this Higgsfield link, and the plan you pick should match the models you actually need.

If your team makes ads every week, the real problem is rarely the chat window. It is where briefs live, who approves what, and where final files are kept. Blur Studio builds the repeatable workflow, approval portal or shared asset library around these tools, so the chat is just one step in a process people can follow.