HeyGen Adds Website-to-Video and 30-Minute Avatars
HeyGen's July 2026 release, published on 17 August 2026, added Video Podcast, code-built motion graphics, website-to-video and 30-minute avatar renders.
HeyGen published its July 2026 product round-up on 17 August 2026, detailing a batch of releases that push the platform well beyond its original avatar-video remit [1]. The headline additions are a Video Podcast tool that turns documents and URLs into two-host video shows, HyperFrames motion graphics generated as code inside HeyGen's Video Agent, 30-minute continuous talking-avatar renders, and two pipelines that generate video directly from existing assets: website-to-video and Figma-to-video [1].
Taken together, the release describes a shift in what HeyGen is selling. The company has historically competed on avatar quality and language coverage; this update competes on how much of a video production process can run without a human assembling it.
What HeyGen shipped in July 2026
Video Podcast
HeyGen's Video Podcast tool converts a script, URL, PDF, document or bare topic into a two-host video podcast, complete with a studio scene, multi-camera cuts and B-roll, which the company says renders within minutes [1]. The AI hosts share a single studio environment and react to one another, with camera coverage automatically alternating between wide shots and close-ups [1]. Users can edit the script before rendering, with voices updating automatically to match, and can invite a real human co-host by email — HeyGen states the invitee does not need a HeyGen account [1]. The tool is live at app.heygen.com/apps/video-podcast [1].
HyperFrames inside Video Agent
HyperFrames generates custom motion graphics as code rather than assembling them from templates, and is now integrated directly into Video Agent [1]. From a single prompt, HeyGen says the agent produces script, voiceover, avatar presenter, pacing, captions, music and motion graphics together, including custom charts, animated text and assembling diagrams, pulling images from the web or generating media itself where needed [1]. The distinction between code-generated and template-based graphics is the substantive part of this claim: templates constrain output to predefined shapes, whereas generated code can in principle produce arbitrary layouts, though HeyGen has not published examples demonstrating the practical range.
30-minute talking avatar videos
HeyGen extended maximum render length to a single continuous 30-minute video, stating this is "6x the industry ceiling and 10x our previous maximum" while maintaining consistent facial appearance throughout [1]. That comparison figure is HeyGen's own characterisation of competitors' limits rather than an independently compiled benchmark, and readers should treat it accordingly. The capability is available on Avatar III, IV and V, through both the API and the web application [1].
Website-to-video and Figma-to-video
Two pipelines generate video from assets a business already owns. Website-to-video takes a URL, captures the site's screens, fonts and colours, then has the agent write a narrative and add narration and music [1]. HeyGen states the first attempt is free, and the tool is live at hyperframes.dev/website [1]. Figma-to-video connects Figma design files directly to HyperFrames, and HeyGen says it preserves all hex codes, fonts and frames in the output, removing manual animation work; it is installed via the command npx hyperframes@latest skills [1].
Also in the release
HeyGen additionally shipped KeyFrames as open source, described as GSAP-style keyframe animation offering motion control through both code and a visual editor; a Media Library containing "10,000+ music tracks, 75,000+ images, plus sound effects, logos" alongside access to generative models, free with a HeyGen login; and a Storyboard Mode that generates five camera angles with frame sketches for approval before rendering [1].
Why this matters
The Figma-to-video pipeline is the most strategically interesting item here, because it connects HeyGen to where design work already lives rather than asking teams to recreate assets inside HeyGen. The same logic applies to website-to-video. Both reflect a broader 2026 pattern across creative AI tooling — Adobe extended Firefly into music, speech and sound effects on 20 August 2026, and xAI added multi-reference consistency to Imagine Video on 31 July 2026 — in which vendors compete on producing finished, assembled output rather than isolated generated clips.
Who should care
Marketing and content teams already producing repeat video formats — product explainers, launch videos, internal updates — are the primary audience, particularly those with existing Figma design systems that the new pipeline could draw on directly. Teams producing long-form training or educational content should note the 30-minute continuous render specifically, since stitching shorter renders together has historically introduced visible discontinuities. Agencies evaluating whether AI video can absorb junior production work now have a fairly concrete test case in website-to-video, which HeyGen offers free on first use.
Practical implications for buyers and users
Because website-to-video's first attempt is free, it is the lowest-cost way to assess the quality of HeyGen's agentic output before committing to anything. Teams with a Figma design system should test whether the claimed preservation of hex codes, fonts and frames actually holds on a real file, since brand fidelity is the entire value proposition of that pipeline and a near-miss on colour is worse than no automation at all. For long-form use, the 30-minute capability should be tested at full length rather than sampled, as facial consistency claims are most likely to degrade toward the end of a render. Note also that several of these tools live on separate surfaces — app.heygen.com, hyperframes.dev, an npx command — rather than a single interface, which is worth factoring into how a team would actually adopt them.
Limitations, availability and unresolved questions
This is a monthly round-up published on 17 August 2026 covering releases across July 2026, so individual features do not carry separate announcement dates and precise availability timing for each is not established. HeyGen has not published pricing for the new capabilities, nor which subscription tiers gate them, beyond stating that the Media Library is free with a login and website-to-video's first attempt is free. The "6x the industry ceiling" claim for 30-minute renders is unverified and not attributed to a specific competitor comparison. HeyGen also has not detailed how the Video Podcast tool handles factual accuracy when generating a show from a URL or topic, which matters given the output is a two-host discussion presented as informed commentary.
Verdict
This is a substantial release rather than a routine monthly update, and its direction is clear: HeyGen is trying to own the assembly of finished video, not just the avatar inside it. Figma-to-video and website-to-video are the most practically useful additions because they start from assets businesses already have, and the free first attempt on the latter makes it genuinely easy to evaluate. The unverified competitor comparison on render length is the weakest part of the announcement, and the absence of tier and pricing detail across a seven-feature release makes it harder than it should be to work out what a given team would actually be able to use.
The current AI Video & Avatars shortlist
Where this sits in the wider market: our current shortlist for AI Video & Avatars, what each tool is best at and the main caution to check before committing.
| Tool | Best for | Current position | Important caution |
|---|---|---|---|
| Runway Production pick | Cinematic generation and controlled editing | Seedance 2.5 is Runway’s current video model, with 1080p output across text-to-video, image-to-video and video-to-video added in August 2026. Runway now also hosts third-party models, so it increasingly acts as a router rather than a single-model tool. | Gen-3 Alpha Turbo and Gen-4 Aleph were retired on 30 July 2026 and those model IDs now fail; confirm current names before API work. |
| Google Veo High-end generation | Video with audio, references and vertical formats | Veo 3.1 supports text and image inputs, audio-capable output and formats ranging from vertical social clips to higher-resolution production workflows. | Access, resolution and cost vary between Gemini, Flow, Vertex AI and API tiers. |
| Adobe Firefly Creative suite | Brand assets and end-to-end Adobe workflows | Firefly combines Adobe’s own models with partner models, storyboards, editing and brand-oriented production inside a broader creative suite. | Partner models can have different training, rights and credit rules from Adobe models. |
| Kling Creator alternative | Character motion and social video experiments | Kling remains a widely used alternative for text-to-video, image-to-video and creator-focused generation. | Plans, limits and model availability change frequently; confirm the current vendor page before purchasing. |
| Luma Dream Machine Fast ideation | Concept clips and visual iteration | Luma’s Dream Machine and Ray family suit rapid visual exploration and image-to-video workflows. | Plans, limits and model availability change frequently; confirm the current vendor page before purchasing. |
| Pika Social effects | Short-form transformations and creator effects | Pika emphasises approachable effects and short-form creation rather than complex production pipelines. | Plans, limits and model availability change frequently; confirm the current vendor page before purchasing. |
| HeyGen Avatar localisation | Sales, training and multilingual presenter videos | HeyGen combines avatar presentation, translation and localisation workflows for business content. | Test brand pronunciation, lip sync and disclosure requirements in every target market. |
| Synthesia Enterprise avatars | Governed training and internal communications | Synthesia focuses on controlled presenter-video production, templates, languages and enterprise deployment. | Plans, limits and model availability change frequently; confirm the current vendor page before purchasing. |
| Descript Editor pick | Editing real footage with AI assistance | For many teams, editing captured video with transcript-led tools is more reliable than generating every frame from scratch. | Plans, limits and model availability change frequently; confirm the current vendor page before purchasing. |
| Gemini Omni Emerging multimodal | Conversational video creation from mixed media | Google’s Gemini Omni direction combines images, audio, video and text inputs with conversational generation and editing. | Treat newly released capabilities as emerging until they pass your own production tests. |
Related reading
Sources and verification notes
Primary product documentation checked for this update: