HeyGen Video is a scene-generation model built on MiniMax H3 and post-trained by HeyGen. It creates visuals and sound together. HeyGen's avatar engines instead animate an existing look from a script. Source: HeyGen model documentation.
That distinction is the first thing to resolve when searching for “HeyGen video.” Are you trying to generate a new scene, or deliver a presentation through a recurring spokesperson? Those are different production decisions, even when both outputs contain a person speaking.
This guide compares documented workflows, rather than ranking image quality. We have not run a controlled HeyGen Video versus MiniMax H3 benchmark. Documentation checked on October 2, 2026.
What can HeyGen Video generate?
The API model is heygen-video-1:
| Mode | Input | Purpose |
|---|---|---|
text_to_video | Prompt | Generate a scene |
image_to_video | Prompt and image | Start from that exact frame |
reference_to_video | Prompt and references | Guide a newly composed scene |
Documented output is 5–15 seconds at 480p or 768p with generated audio. Reference mode requires an image or video; audio alone is insufficient. Source: HeyGen modes and parameters.
For planning, separate an opening-frame requirement from an identity requirement. A storyboard still might need to establish the opening composition. A product photograph might instead serve as a visual reference while you invent a different setting. Write down which constraint matters before selecting your workflow.
HeyGen Video vs MiniMax H3: what actually differs?
MiniMax H3 is the underlying model family; HeyGen Video is a post-trained offering with its own API contract. Sharing a foundation does not establish identical output or a quality winner.
MiniMax publishes H3-Base FL2VA and Ref2VA weights. Its documented full system adds context processing and a separate 2K regeneration stage; those hosted components are not included in the released base weights. Source: MiniMax H3 model card.
| Decision | What to evaluate |
|---|---|
| Existing HeyGen integration | Whether its model endpoint fits your current application |
| Local experimentation | Whether the released H3 weights and your hardware fit the task |
| Finished delivery size | The actual endpoint output, plus any separate regeneration or editing steps |
| Brand consistency | Whether your own reference assets survive across several candidate clips |
| Production economics | The total spend and editing time required for an approved result |
A model family's capabilities are not a promise that every provider exposes every control. Compare the endpoint you will actually use. If local graph control matters, our MiniMax H3 ComfyUI guide explains that route.
How do you access the HeyGen Video API?
Use a paid API key and submit to POST /v3/models/videos; retrieve status through GET /v3/models/videos/{video_id}. Creation is asynchronous. Completed jobs provide a signed download URL. Source: HeyGen API workflow.
Before integrating, define what your application should do while a clip is queued, when generation fails, and after the result arrives. Save approved output into your normal media workflow instead of treating a temporary delivery link as your permanent publishing asset.
What does HeyGen Video cost?
We could not verify a public, model-specific rate from the linked pricing screen during this review. Check HeyGen's API pricing and your account's billing terms before budgeting. Do not assume an avatar subscription or another video model's advertised rate applies here.
For a useful comparison, calculate:
Cost per approved clip = total generation spend ÷ approved clips.
Then track editing time separately. For example, a hypothetical batch costing $12 with three approved clips costs $4 per approved clip before editing. This is an illustration, not a HeyGen or MiniMax price quote. A cheaper attempt can still be expensive if most outputs need replacing.
How to test a product-video workflow fairly
Use a small brief that represents a real deliverable: one product, one setting, one camera plan, and one clearly defined action. Decide what counts as a usable result before generating anything.
- Use the same source photograph and creative objective in both workflows.
- Match duration and delivery size where the endpoints permit it.
- Keep a record of prompt changes and any automatic enhancement settings.
- Generate multiple candidates within a fixed budget.
- Review shape, color, markings, motion, sound, and the final frame.
- Record rejected clips and editing effort alongside successful outputs.
Do not assume matching seed numbers make outputs comparable across different models. The goal is a comparable brief and review standard, not identical pixels.
HeyGen flags long text, delicate organic motion and close hand manipulation as weaker areas. Source: HeyGen limitations.
For your own review, inspect any readable packaging frame by frame. Add final pricing, disclosures, and legal copy in an editor when exact typography is essential. A clip can look convincing at normal playback speed while still containing a label error that makes it unusable.
Which workflow should you choose?
Start with the deliverable. For a repeated spokesperson presentation, evaluate an avatar workflow. For an invented environment, product scene, or short visual concept, evaluate scene generation. For experiments that require inspecting and changing a local graph, consider the H3 local route.
If you want to explore the foundation through this website, try MiniMax H3 online or browse MiniMax H3 prompt examples. H3Vid offers a separate MiniMax H3 workflow; it does not currently offer HeyGen Video. Choose between them using your own approved outputs and actual billing, rather than assuming the post-trained version must be better for every brief.