PixVerse V6 vs C1: What’s Different, Which to Use, and How to Call the API (2026)

PixVerse released two video models eight days apart this spring, and the names don’t tell you which one to pick. Here’s the split, with Gateway pricing and working API calls.

PixVerse V6 arrived on March 30, 2026 as the successor to V5.6: variable clip length up to 15 seconds, native audio, and a set of cinematic camera controls. PixVerse C1 followed on April 7, positioned by the company as its “first AI video model for film production,” tuned for fast motion, combat choreography, and effects-heavy scenes. Both top out at 1080p and 15 seconds with synchronized audio, both support text-to-video, image-to-video, and reference-to-video, and both are now live on Yotta AI Gateway. So the decision isn’t about specs. It’s about what you’re shooting.

The two models side by side

Two things in that table drive most decisions. V6 has the deeper control surface: if you know what a 35mm lens at f/2 looks like, V6 lets you ask for it, and it’s the only one of the two with a multi-shot flag and reference videos. C1 trades that for motion quality; PixVerse built it for the shots that break other models, fight scenes, fast camera moves, particles, fluid, and it segments a longer prompt into shots without being told to.

What “reference-to-video” means, and why both models have it

Image-to-video animates one picture: the frame you upload becomes the first frame of the clip. Reference-to-video is different. You upload one or more reference images, a character, a product, a wardrobe piece, and the model generates a new scene from scratch that keeps those references consistent. The reference doesn’t appear as a frame; it appears as an identity the model holds onto while it invents the shot.

That’s the mode that matters for anything branded. A product that has to look like the product, a character that has to look the same across five clips, a garment that has to keep its pattern.

On the Gateway, reference-to-video has its own endpoint. You pass a reference_items list, each item a URL, a type, and a ref_name, and you refer to each one in the prompt as @ref_name followed by a space (PixVerse requires the space; @Image1. or @Image1, fails). V6 takes 1 to 10 images and up to 2 reference videos, with the total video length capped at 15 seconds; a request that includes reference videos costs double and needs aspect_ratio set to auto. C1 takes 1 to 7 images and no videos. Images are jpeg, jpg, or png, up to 20 MB each.

Both PixVerse models support it, alongside Seedance 2.5 and 2.0, Wan 3.0 and 2.7, and HappyHorse-1.0, which is the whole reference-to-video lineup on the Gateway today.

Price: tiered by resolution, and audio costs extra

PixVerse bills per second of output, and the rate depends on resolution and on whether you ask for audio. Current Gateway rates:

Same table for every mode on each model: text-to-video, image-to-video, and reference-to-video all bill at these rates, with the one exception that a reference-to-video request with reference videos costs double.

In real clips: a 5-second 720p clip with audio, the default resolution, is about 20 cents on V6 and 22 cents on C1. A 15-second 1080p clip with audio is $1.17 on V6 and $1.22 on C1. The difference between the two models is small at every tier, about a third of a cent per second at 1080p, so price shouldn’t be the reason you pick one over the other.

Price is the reason you pick PixVerse over the rest of the catalog. The same 15-second 1080p clip runs $6.45 on Seedance 2.5 ($0.43 per second at 1080p) and $2.10 on Wan 3.0 ($0.14 per second at 1080p). At 720p the spread is similar: $0.25 per second on Seedance 2.5 against about $0.04 on PixVerse V6 with audio. That matters more in video than in text because iteration is the workflow. Nobody keeps the first generation. A pipeline that runs five takes per shot on PixVerse costs about what one take costs on Seedance 2.5.

Which one to use

Choose V6 if the shot is dialogue, product, narrative, or anything where you want to specify the camera. It’s the general-purpose model, it has the multi-shot flag, it accepts reference videos, and it’s slightly cheaper at every tier.

Choose C1 if the shot has fast motion, contact, effects, or a crowd. That’s what it was trained for, and V6’s own documentation lists chaotic action and complex choreography as the cases that need retries.

For a lot of pipelines the answer is both: V6 for the establishing and product shots, C1 for the action beat in the middle, same API, same key, different model string.

How to call the PixVerse API on Yotta AI Gateway

All video generation on the Gateway is asynchronous: submit a request, get a request ID, poll until the status is completed, then read the output URL. The base URL for video is https://gateway.yottalabs.ai/api/maas, every call authenticates with your Yotta key in the X-API-KEY header, and each mode has its own endpoint. The model strings are pixverse-v6-t2v, pixverse-v6-i2v, pixverse-v6-r2v, pixverse-c1-t2v, pixverse-c1-i2v, and pixverse-c1-r2v.

Text-to-video, PixVerse V6

curl -X POST "https://gateway.yottalabs.ai/api/maas/text-to-video/generations" \
  -H "X-API-KEY: $YOTTA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "pixverse-v6-t2v",
    "parameters": {
      "prompt": "Slow dolly-in on a ceramic coffee mug on a wooden counter, morning window light, shallow depth of field, 35mm lens. Audio: quiet room tone, a distant kettle.",
      "aspect_ratio": "16:9",
      "resolution": "1080p",
      "duration": 8,
      "generate_audio": true
    }
  }'

aspect_ratio accepts 16:9, 4:3, 1:1, 3:4, 9:16, 2:3, 3:2, and 21:9. duration is 1 to 15 seconds; resolution defaults to 720p. Add "multi_shot": true on V6 to get several shots in one clip, and "seed" for repeatable output. C1’s text-to-video takes the same parameters minus multi_shot.

Image-to-video, PixVerse C1

curl -X POST "https://gateway.yottalabs.ai/api/maas/image-to-video/generations" \
  -H "X-API-KEY: $YOTTA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "pixverse-c1-i2v",
    "parameters": {
      "prompt": "The sprinter explodes off the blocks, camera tracking alongside at ground level, dust and light flares, high-speed motion.",
      "first_frame": "https://your-cdn.example.com/sprinter-blocks.png",
      "resolution": "1080p",
      "duration": 6,
      "generate_audio": true
    }
  }'

first_frame is a URL to a jpeg, jpg, png, or webp up to 20 MB. Add last_frame and the request becomes a first-to-last-frame transition. There’s no aspect_ratio here; the clip follows the image.

Reference-to-video, PixVerse V6

curl -X POST "https://gateway.yottalabs.ai/api/maas/reference-to-video/generations" \
  -H "X-API-KEY: $YOTTA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "pixverse-v6-r2v",
    "parameters": {
      "prompt": "@Image1 walks through a rainy Tokyo street at night wearing @Image2 , neon reflections on wet pavement, handheld camera.",
      "reference_items": [
        {"url": "https://your-cdn.example.com/character.png", "type": "image", "ref_name": "Image1"},
        {"url": "https://your-cdn.example.com/jacket.png", "type": "image", "ref_name": "Image2"}
      ],
      "aspect_ratio": "9:16",
      "resolution": "1080p",
      "duration": 8,
      "generate_audio": true
    }
  }'

Every ref_name has to be unique, 30 characters or fewer, with no special characters, and each @ref_name in the prompt needs a space after it. To add a reference video on V6, include an item with "type": "video" (mp4, mov, or webm, up to 50 MB) and set aspect_ratio to auto. Swap the model string to pixverse-c1-r2v for C1, with images only.

Poll for the result

curl "https://gateway.yottalabs.ai/api/maas/reference-to-video/generations/$REQUEST_ID" \
  -H "X-API-KEY: $YOTTA_API_KEY"

The submit call returns data.request_id. The poll endpoint matches the mode you submitted to (/text-to-video/generations/{id}, /image-to-video/generations/{id}, or /reference-to-video/generations/{id}) and returns status, then output_url and duration once the status is completed. The Seedance API guide has a full Python client with the polling loop that works for every video model on the Gateway; swap the model string and it runs PixVerse.

Frequently asked questions

What’s the difference between PixVerse V6 and C1? V6 is the general cinematic model with 20-plus camera controls, a multi-shot flag, and reference-video support. C1 is built for action, combat, VFX, and fast motion, with automatic shot segmentation from the prompt. Same 15-second, 1080p, native-audio ceiling on both.

Which is cheaper? V6, by a little, at every tier: $0.0170 per second at 360p against $0.0204 for C1, and $0.0782 against $0.0816 at 1080p with audio. Both are far under Seedance 2.5, which is $0.43 per second at 1080p on the Gateway.

Does PixVerse support reference-to-video? Yes, both V6 and C1 do. You pass a reference_items list of image URLs with a ref_name each and refer to them in the prompt as @ref_name. V6 takes up to 10 images and 2 reference videos; C1 takes up to 7 images.

How long can a PixVerse video be? 1 to 15 seconds per request on both models. V6 also supports multi-shot generation within one request.

Does PixVerse generate audio? Yes. Set generate_audio to true and both models produce synchronized audio in the same pass, at a higher per-second rate. Multi-character dialogue is the documented weak spot on V6; expect retries there.

PixVerse V6 vs Seedance 2.5? Seedance 2.5 goes to 30 seconds per request and accepts far more reference material; PixVerse V6 caps at 15 seconds but costs about a fifth as much per second at 1080p. Seedance for long-form and heavy reference work, PixVerse for volume. The Seedance 2.5 vs 2.0 breakdown covers the ByteDance side.

Is PixVerse on Yotta AI Gateway? Yes. PixVerse V6 and C1 are both live in text-to-video, image-to-video, and reference-to-video modes, alongside Seedance 2.5 and 2.0, Wan 3.0 and 2.7, HappyHorse-1.0, and Kling v3, all behind one API key.

Bottom line

PixVerse V6 is the low-cost way to generate current-generation video with real camera control, and C1 is the model you switch to when the shot moves fast. Same ceiling, same API, a third of a cent apart per second. Run both through Yotta AI Gateway with one key, and the choice per shot is a model string. The rest of the video lineup is in the Gateway announcement.