Text assistants like ChatGPT and Claude get most of the attention, but some of the biggest practical changes in 2026 are in voice, video, and 3D. A small business can now run a phone agent that answers in natural speech, produce a short product video without a film crew, or turn a photo into a 3D model in under a minute.
At Signal & Form, Kavan Sohal pays for and uses each of these tools, on client voice-agent builds and on new web properties, so this guide comes from hands-on use on paid plans, not free trials. Every product fact below was checked against the vendor's own pages on October 7, 2026. Prices are in US dollars unless noted and change often.
Key Takeaway
The question to ask of every tool in this category is not "how good is the output?" It is "am I allowed to use this output commercially, and did everyone whose voice or face appears in it consent?" Get that wrong and the time savings do not matter.
ElevenLabs: AI voice and voice agents
ElevenLabs is an AI voice platform covering text-to-speech, voice cloning, dubbing, speech-to-text, and ElevenAgents, a platform for building voice agents that combines speech recognition, a language model of your choice, and its own voices. Agents can answer phone calls through SIP trunking or a native Twilio integration, and run batch outbound calls.
Source: ElevenLabs, ElevenAgents overview
Its newest voice models, Eleven v4 and Eleven v4 Turbo, launched in late September 2026 with support for more than 90 languages. Turbo is built for real-time conversation.
Source: ElevenLabs, Eleven v4
Pricing: the free plan includes 10,000 credits a month, about 10 minutes of speech. Paid plans start at Starter, US$6 a month for 30,000 credits, and Creator, US$22 a month. Voice agents cost US$0.08 per minute on every plan after the included minutes, with language-model usage billed on top and phone carrier charges billed by your telephony provider.
Source: ElevenLabs, pricing
Source: ElevenLabs, agent pricing
Rules to know: the free plan is for non-commercial use only. Instant voice cloning starts on Starter. You may only clone your own voice or a voice you have permission to use, and ElevenLabs' usage policy requires voice agents to tell callers they are talking to an AI. By default, data is stored in the United States; regional data residency is an Enterprise feature and does not currently include a Canadian region.
Source: ElevenLabs, terms of use
Source: ElevenLabs, prohibited use policy
Source: ElevenLabs, data residency
OpenAI realtime voice: the developer route
For businesses with a developer, OpenAI's realtime models take speech in and produce speech out directly, which keeps conversations fast and natural. They connect over WebRTC or WebSocket for web and app agents and over SIP for phone calls. The current models are gpt-realtime-2.1 and the lower-cost gpt-realtime-2.1-mini.
Source: OpenAI, Realtime API guide
Source: OpenAI, API changelog
Pricing: pay as you go, billed per million audio tokens. There is no free tier. Check OpenAI's pricing page for current rates.
Rules to know: OpenAI does not use API data to train its models by default, and keeps abuse-monitoring logs for up to 30 days. Its usage policies require a clear disclosure that the voice is AI-generated. Custom voices are limited to approved organizations and require a recorded consent statement from the speaker.
Source: OpenAI, how your API data is used
Source: OpenAI, text-to-speech guide
ElevenLabs or OpenAI? ElevenLabs is the faster route to a working agent with a wide voice library and a no-code builder. OpenAI realtime suits teams that want to build the agent into their own application. Either way, a well-built agent confirms details with the caller before booking anything, logs every call, and hands off to a person when it should. Our AI automation work covers these builds.
Seedance: AI video from ByteDance
Seedance is ByteDance's video generation model. It takes text, image, audio, and video references and produces video with synchronized sound. The latest version, Seedance 2.5, reached international customers through ByteDance's BytePlus cloud in August 2026, with clips up to 30 seconds.
Source: BytePlus, Seedance 2.5
How to access it: through the BytePlus ModelArk API, which BytePlus lists as available in Canada; through partner platforms including Higgsfield and Runway; or through ByteDance's Dreamina and CapCut apps.
Source: BytePlus, ModelArk availability
Rules to know: Seedance output carries visible and invisible watermarks and C2PA content credentials, and BytePlus restricts generating video from images of real people's faces without verification. The model has also been controversial: in February 2026, Disney sent ByteDance a cease-and-desist letter over Seedance generating its characters. Keep prompts away from other companies' characters, logos, and real people.
Source: Axios, Disney cease-and-desist to ByteDance (February 2026)
Higgsfield: many video and image models in one subscription
Higgsfield is a subscription platform that bundles many AI image, video, and audio models, including Seedance, Kling, Veo, Sora, and its own models, so you can switch between them without separate accounts.
Source: Higgsfield, which model should I use
Pricing: the free tier covers a small set of models without credits, and paid plans include monthly credits that do not roll over. Check Higgsfield's pricing page for current plans.
Source: Higgsfield, how plans work
Rules to know: Higgsfield says you own your outputs and does not tie commercial use to a plan, but watermark-free downloads require a paid plan. Importantly, Higgsfield may use your uploads and generations to train its models on every plan except Enterprise, including paid ones. Do not upload anything confidential, and do not upload client material without permission.
Source: Higgsfield, ownership and commercial use
Tripo: text and images to 3D models
Tripo generates 3D models from a text description or an image, with automatic rigging and animation presets, and exports standard formats such as GLB and FBX. Its current models include Tripo H3.1 for detailed, textured models and Tripo P1 for low-poly game assets.
Source: Tripo, models
Uses for a business: product models for an online store or configurator, assets for games and interactive marketing, and quick prototypes for packaging or retail displays.
Rules to know: Tripo's paid plans give private models and commercial rights. Treat the free tier as non-commercial: Tripo's terms say it retains rights to free users' outputs. Check Tripo's pricing page for current plans.
Source: Tripo, terms of service
The Canadian rules that apply
There is no Canadian law written specifically for commercial voice cloning or AI likeness, so general rules apply. A few that matter most:
- Privacy: Canada's federal, provincial, and territorial privacy commissioners have published principles for generative AI that call for meaningful consent, no deceptive design, and clear identification of AI-generated content in high-impact uses.
- Advertising: the Competition Bureau has said the Competition Act's rules on misleading representations apply to AI-generated ads and fake endorsements, and misleading telemarketing, including recorded and automated calls, can be a criminal offence.
- Deepfakes: the Protecting Victims Act (Bill C-16), in force since July 2026, makes distributing sexual deepfakes without consent a criminal offence.
Source: Office of the Privacy Commissioner of Canada, generative AI principles
Source: Competition Bureau, synthetic media in the digital landscape
Source: Department of Justice, Bill C-16
This is general information, not legal advice. Before you clone a voice, use a real person's likeness, or put an AI agent on the phone with customers, have counsel review the use, and get written consent from anyone whose voice or face you use.
A practical checklist before you start
- Use a paid plan for anything commercial, and keep the licence terms on file.
- Get written consent from anyone whose voice or likeness you use, including staff.
- Tell people when they are talking to or watching AI.
- Keep confidential and client material out of tools that train on uploads.
- Check where each vendor stores data if your industry or clients have residency requirements.
- Keep a person reviewing anything published or sent to customers.
Key Takeaway
Start with one contained use, such as a voice agent for after-hours calls or a short product video, measure the result, and expand from there.
If you want help choosing and building, see how Kavan approaches AI consulting in Canada, or read our guide to agents, skills, and connectors.
Thinking about a voice agent or AI video for your business? Book a free call and we will tell you what is realistic, what it costs, and what to set up first.
Get Started