HeyGen Video 1
HeyGen Video 1 is the #1 video model on OpenRouter this week: 57K requests, ahead of Veo 3.1 Lite (50K) and Seedance 2.5 (29K) (OpenRouter video rankings, weekly requests, October 9, 2026). It's live on varg as heygen_video_1.
What makes it different: one call gives you the picture and the sound together. You can turn a product still into a 10-second shot with the engine roaring and the tunnel echoing, or write a scene and get the dialogue, room ambience and effects with it.
What HeyGen Video 1 does
- Text to video — describe a scene and get a clip with sound.
- Image to video — give it a first frame and it animates from that exact image.
- Reference to video — pass several image, video or audio references to keep a product, person or place consistent.
- 5–15 seconds per clip, landscape, square or portrait, up to 2K.
- Sound in the same call — dialogue, ambience and sound effects; no separate lip-sync step.
It's built on MiniMax H3 and post-trained by HeyGen for "production-quality video without production-level costs" (HeyGen).
How good is it
HeyGen published a blind preference test (650 votes) against H3 Max, H3 Max Turbo, Seedance 2.0, Kling 3.0 Pro and Veo 3.1. Overall, 55.8% preferred HeyGen Video. Normalised Elo: HeyGen Video 1000, Seedance 2.0 955, H3 Max 952, Kling 3.0 Pro 857, Veo 3.1 744.
| Where it wins | Preferred | Where it trails | Preferred |
|---|---|---|---|
| Timing | 66% | Consistency | 43% |
| True to the source image | 65% | Music | 45% |
| Natural motion | 63% | Detail | 46% |
| Follows the prompt | 62% | Camera motion | 47% |
| Physics | 59% | ||
| Sound & voices | 57% |
Source: HeyGen Video 1.0 catalog. It's also fast: 3.7 s of model inference for a 10-second image-to-video clip, against 8.3 s for H3 Max.
What that means in practice: it's strongest at short, contained shots — one subject, one action, real-feeling physics, and sound that matches. It's weaker at keeping a subject identical across a long, moving shot.
We tested it: one still → a 10-second clip with sound
In the varg playground we took one frame from our launch poster — a classic car in a tunnel, with the text cropped out — pasted it as the first frame, and wrote:
The classic car accelerates through the tunnel, engine roaring, tunnel lights streaking past, tires hissing on the wet road.
16:9, 10 seconds. It came back in 2K with the engine, the tunnel echo and the tire hiss, billed 95 credits. It starts from our exact frame — then, as the car drives away, its shape drifts a little: by the last second it reads as a slightly different car. For product shots, keep clips short or pass more references.
Every clip in our walkthrough video, including the headphones in the intro, was generated by HeyGen Video 1.
How to prompt HeyGen Video 1
- Name the camera, not the mood. "Cinematic" does almost nothing. "Shot on an ARRI Alexa Mini, 50mm at f/2, fine film grain, 24 fps" changes depth of field, grain and motion together.
- One subject, one action. Don't pack several locations and a storyline into a 10-second clip.
- Direct the sound. Audio comes from the same prompt — name the room tone, the effects and how far away they are. "Sound: a deep bass impact, then glass falling on concrete."
- Keep the look honest. HeyGen's own prompts end with "muted true-to-life colour, no teal-orange grade, no CG sheen."
- Control lettering. Words on screen are drawn by the model inside the shot. Keep them short and say they're "the only lettering in frame"; otherwise add "no logos, brand names or printed words anywhere in frame."
Pricing
| Price | |
|---|---|
| HeyGen direct (768p, with audio) | $0.02/s ($0.01/s launch promo through October) |
| varg, 480p | ~2 credits/s |
| varg, 720p | ~3 credits/s |
| varg, 2K | ~9 credits/s |
1 varg credit = $0.01. Our 10 s 2K image-to-video run: 95 credits. For comparison, HeyGen lists Veo 3.1 at $0.40/s and Seedance 2.0 at $0.303/s at 768p with audio.
Code example
Image to video with the varg API:
curl -X POST https://api.varg.ai/v2/video \
-H "Authorization: Bearer $VARG_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "heygen_video_1",
"prompt": "The classic car accelerates through the tunnel, engine roaring, tunnel lights streaking past, tires hissing on the wet road.",
"files": ["https://your-cdn.com/first-frame.png"],
"duration": 10,
"aspect_ratio": "16:9"
}'
Drop files for text to video. Poll GET https://api.varg.ai/v2/jobs/{id} until status is completed; the clip URL is in output. To see every input the model accepts: GET https://api.varg.ai/v2/tools/video?model=heygen_video_1.
Or run it without code in the varg playground.
When to use HeyGen Video 1 vs. other models on varg
- HeyGen Video 1 — short shots where sound matters and the budget is tight: product moments, ad cutaways, explainer b-roll, scenes with a spoken line.
- Seedance 2.5 / Veo 3.1 — when you need more detail or longer, more complex camera moves, and can pay more per second.
- HeyGen Avatar — a talking presenter from a script, not a scene.
FAQ
What is HeyGen Video 1?
HeyGen's first general-purpose video model: text, a first frame or references in, a 5–15 second clip with sound out. Built on MiniMax H3, post-trained by HeyGen.
How much does HeyGen Video 1 cost via varg?
About 2 credits per second at 480p, 3 at 720p and 9 at 2K (1 credit = $0.01). A 10-second 2K clip cost us 95 credits.
Does HeyGen Video 1 generate audio?
Yes — dialogue, ambience and effects, in the same call as the picture. Describe them in the prompt.
Do I need a HeyGen account?
No. Use your varg API key or the varg playground.
What is HeyGen Video 1 weakest at?
Consistency, fine detail, camera motion and music, per HeyGen's blind tests. In our test the car's shape drifted over 10 seconds.