Item one: HeyGen Video arrives in Creative Studio
Until October 21, HeyGen Video lived in the API. Now it sits inside Creative Studio, the app that replaced the old video generator and image generator. It has three tabs: video, image and voice.
A clip runs 5 to 15 seconds, resolution goes from 480p to 2K, and the audio is generated in the same pass as the picture. Input comes in three shapes: text, a first frame, or a reference set. A reference set holds up to nine images, three videos and three audio files, addressed in the prompt by name, for example Picture 1.
In the API it is one request, then a webhook or a poll on GET /v3/videos/{video_id}. Output is MP4 in H.264 at 24 frames per second with stereo AAC. A seed can be fixed, so the same prompt returns the same clip.
What a second of video costs in credits
Creative Studio shows a rate next to every model in the picker. HeyGen Video charges 0.2 credits per second at 480p, 0.24 at 768p and 0.67 at 1080p and 2K. Length is limited to 5, 8, 10 or 15 seconds.
For comparison, Seedance 2.5 in the same list costs 7.7 credits per second at 720p and 18.9 at 1080p, Veo 3.1 takes 90 credits per video of any length, Wan 3 takes 17, LTX 2 Fast takes 15. A fortyfold spread per second explains why the model picker became the main decision in the interface.
One detail from the documentation matters for budgets: 1080p and 2K bill at three times the 768p rate. A five-second 2K clip costs 3.35 credits, the same clip at 768p costs 1.2.
What changed in the interface itself
Creative Studio builds a scene out of shots: a prompt, a model, a length, a resolution, an aspect ratio, a credit estimate and a Generate button. Inspiration tiles under the composer fill in a prompt and settings for you.
Plans are strict about access. The free tier locks every video model, Creator opens almost everything except Seedance 2 and 2.5, and Pro and above opens all of it. Enterprise accounts stay on the classic video and image generators for now.
Prompt enhancement comes in four modes: Turbo, quality, ultraquality, or off when the model must follow the text word for word. Aspect ratios cover 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16, with an adaptive mode for references. A failed generation refunds its credits automatically, and a Re-edit button restores the prompt, references, model and settings in one click.
Why it matters
AI video production used to hinge on two things: the price of a usable take and the separate job of stitching sound onto it. The first is now a credit rate per second, the second disappears because dialogue, ambience and effects arrive from the same prompt.
For vertical video ads the arithmetic is simple. An eight-second 1080p clip costs 5.36 credits, so twenty versions of one shot cost 107 credits, which is less than a single Veo 3.1 video.
Item two: the voice is free until November 30
The second release that day concerns sound on its own. HeyGen Voice Turbo, the fast mode of the company's speech model, is open in the API at no charge until November 30, 2026. The price is listed as zero dollars per million characters, no card required, and a key works immediately.
After November 30, Turbo costs four dollars per million characters. The full HeyGen Voice model costs fifteen dollars per million characters and is required for professional voice clones. Public voices and instant clones work in the free mode.
What "free" means, and where the traps sit
The free mode runs on a best-effort basis: no uptime or latency guarantee, and when capacity is full the request returns 429 and has to be retried later. That is a normal arrangement for prototypes and draft narration, and a risky one for anything live or for a pipeline that bills by the minute. A 429 is solved with a queue and a retry, which is exactly the case where free costs engineering time.
Professional voice clones are not available in the free mode, they need the paid heygen-voice-1 model. Creating an instant clone from one recording is free during the preview, and a preview is a preview.
The ranking HeyGen cites comes from the Artificial Analysis leaderboard, captured on October 20. Turbo sits first, HeyGen Voice second, followed by Qwen-Audio-3.1, Eleven v4 Turbo and Eleven v4. Every model reads the same eight cloned voices, four US and four UK.
Why it matters
Speech synthesis used to cost either money or quality. A free million characters is roughly a hundred minutes of speech, enough to test ten languages or to build a scratch track for short-form video at scale.
Batches of vertical clips benefit first, because the voice changes from version to version and recording a narrator does not scale. Channels that count minutes in the hundreds come next, and after November 30 the meter will be visible.
What this changes for a client
First, a project budget now runs in two currencies: credits per second of picture and characters of speech. Second, the free voice removes the need to buy a separate narration service during testing. Third, 2K stopped being a premium tier, it simply costs three times 768p, which is more honest than a hidden surcharge.
Compared with the earlier launch, a HeyGen Video 1 MCP setup solved the access problem. Today's release solves price and assembly. The difference is that the model used to stand alone, and now it is a storefront of a dozen models with a visible rate card.
---
