Video Generation Costs: Per-Second vs Per-Call
Break down video generation pricing for 2026: per-second, per-call, and per-token models across MiniMax H3, Kling 3.0, and Gemini Omni Flash with budget templates.
TL;DR — Video generation APIs price in three different ways: per-second of output (Kling: $0.02-$0.10/sec), per-call flat rate (MiniMax H3: $0.20-$0.80/video), and per-token (Gemini Omni Flash: token-based with variable cost). A 10-second video ranges from $0.20 to $1.00 depending on model and tier. This guide gives budget templates for 10, 100, and 1,000 videos/month.
You want your agent to generate videos. The pricing pages say “$X per second” or “$Y per generation” — but which model is actually cheapest for your use case? The answer depends on video length, resolution, and quality tier.
Let me save you the spreadsheet work.
Three pricing models
| Model family | Pricing basis | What you pay for | Predictability |
|---|---|---|---|
| Kling Video 3.0 | Per-second × quality tier | Length + resolution + pro/turbo | Predictable (linear with duration) |
| MiniMax H3 | Per-call (flat or tiered) | Fixed per video regardless of length* | Most predictable |
| Gemini Omni Flash | Token-based | Input prompt tokens + output estimation | Least predictable |
*MiniMax H3 charges per generation with length tiers; a 5-second and 10-second video may have different rates.
Kling Video 3.0: per-second pricing
Kling offers the most granular control with multiple tiers:
| Tier | Resolution | Quality | Per-second cost | 5s video | 10s video |
|---|---|---|---|---|---|
| Turbo Standard | 720p | Standard | $0.02/sec | $0.10 | $0.20 |
| Turbo Pro | 1080p | Standard | $0.04/sec | $0.20 | $0.40 |
| Omni Pro | 1080p | High | $0.07/sec | $0.35 | $0.70 |
| Omni Pro (4K) | 4K | Highest | $0.10/sec | $0.50 | $1.00 |
Prices estimated from official documentation as of July 2026. Check current rates in SandBase’s model catalog.
Kling cost by use case
| Use case | Typical length | Recommended tier | Cost/video | 100 videos/month |
|---|---|---|---|---|
| Social media clips | 5 seconds | Turbo Standard | $0.10 | $10 |
| Product demos | 10 seconds | Turbo Pro | $0.40 | $40 |
| Ad creative | 10 seconds | Omni Pro | $0.70 | $70 |
| Premium content | 15 seconds | Omni Pro | $1.05 | $105 |
The turbo vs omni decision
Turbo is 2-3× faster to generate (relevant for real-time agents) but lower quality. Omni Pro produces cinema-quality output but takes 60-120 seconds per generation.
For an agent producing social media clips at scale: Turbo Standard. You’re paying $0.02/sec for 720p, which is more than adequate for vertical short-form video.
For an agent creating hero product videos: Omni Pro. The $0.07/sec buys you quality that doesn’t look AI-generated to a casual viewer.
MiniMax H3: per-call pricing
MiniMax H3 is the newest entrant (July 2026). Its standout feature: native 2K stereo video with audio.
| Operation | What it does | Estimated cost | Output |
|---|---|---|---|
minimax/h3/text-to-video | Text prompt → video | $0.30-$0.80 | 2K video with stereo audio |
minimax/h3/image-to-video | Image + prompt → video | $0.30-$0.80 | Animated from reference image |
minimax/h3/reference-to-video | Reference video + prompt → new video | $0.50-$1.00 | Style/motion transfer |
Pricing is per-generation. Exact rates depend on duration and resolution settings. Check SandBase model catalog for current pricing.
H3’s unique value: audio included
Other video models generate silent video. You then need a separate TTS or music generation step. MiniMax H3 produces synchronized stereo audio in the same generation — no additional API call, no audio syncing headache.
For an agent that produces complete video assets (social posts, ads, product clips): this eliminates a workflow step. The cost comparison should account for the audio generation you are not doing separately.
H3 cost math
| Scenario | Videos/month | Avg cost/video | Monthly | Without audio (add ~$0.05) |
|---|---|---|---|---|
| Social content pipeline | 100 | $0.40 | $40 | $45 (video + separate audio) |
| Ad creative testing | 50 | $0.60 | $30 | $32.50 |
| Premium product videos | 20 | $0.80 | $16 | $17 |
The audio inclusion saves 10-15% vs doing video + audio separately.
Gemini Omni Flash: token-based
Gemini’s video generation prices differently — as token consumption rather than per-second or per-call.
This makes cost estimation harder because output tokens vary with video complexity. Approximate ranges:
| Input | Output (estimated) | Approximate cost |
|---|---|---|
| Short text prompt (50 tokens) | ~5,000-10,000 tokens equivalent | $0.10-$0.20 |
| Image + text prompt | ~8,000-15,000 tokens equivalent | $0.15-$0.30 |
| Reference video + prompt | ~15,000-30,000 tokens equivalent | $0.30-$0.60 |
Gemini’s advantage: it’s fast. Generation time is significantly shorter than Kling Omni or H3. For latency-sensitive agents that need a video now (real-time content generation), the speed premium may justify the price uncertainty.
Budget templates
10 videos/month (testing / occasional use)
| Model | Configuration | Monthly cost |
|---|---|---|
| Kling Turbo Standard | 5s, 720p | $1.00 |
| Kling Omni Pro | 10s, 1080p | $7.00 |
| MiniMax H3 | Text-to-video, standard | $4.00 |
| Gemini Omni Flash | Short prompts | $1.50 |
100 videos/month (production content pipeline)
| Model | Configuration | Monthly cost | Cost/video |
|---|---|---|---|
| Kling Turbo Standard | 5s, 720p | $10 | $0.10 |
| Kling Turbo Pro | 10s, 1080p | $40 | $0.40 |
| MiniMax H3 | Mixed, with audio | $50 | $0.50 |
| Gemini Omni Flash | Mixed prompts | $20 | $0.20 |
1,000 videos/month (scale / A-B testing)
| Model | Configuration | Monthly cost | Cost/video |
|---|---|---|---|
| Kling Turbo Standard | 5s, 720p | $100 | $0.10 |
| Kling Omni Pro | 10s, 1080p | $700 | $0.70 |
| MiniMax H3 | Mix of operations | $500 | $0.50 |
At 1,000 videos/month, the tier choice matters enormously: $100 vs $700. The quality-cost trade-off decision should be made with actual A/B test data on your audience.
Decision framework
| If you need… | Choose | Why |
|---|---|---|
| Cheapest possible video | Kling Turbo Standard | $0.02/sec, 720p good enough for social |
| Best quality (cinema-grade) | Kling Omni Pro | Higher cost but highest fidelity |
| Video with audio (no extra step) | MiniMax H3 | Native stereo audio, single generation |
| Fastest generation time | Gemini Omni Flash | Token-based pricing, fast inference |
| Reference-style transfer | MiniMax H3 ref-to-video | Style/motion transfer from existing video |
| Agent needs real-time output | Gemini Omni Flash or Kling Turbo | Both optimize for speed over quality |
The hidden costs
Video generation has costs beyond the API call:
| Hidden cost | Impact | Mitigation |
|---|---|---|
| Failed generations | 5-15% of calls produce unusable output | Budget 1.1-1.2× expected volume |
| Prompt iteration | First prompt rarely produces perfect results | Allocate 3-5 test calls per final video |
| Storage | 2K videos are 5-20MB each | Object storage at ~$0.023/GB/month |
| Review time | Human review for brand safety | Batch reviews, not per-video |
For a production pipeline of 100 videos/month, realistic total cost:
API cost: $40 (Kling Turbo Pro)
Failed retries: $6 (15% fail rate)
Prompt testing: $4 (2 iterations avg)
Storage: $0.50 (500MB/month)
────────────────────────────
Total: ~$50.50/month (vs $40 raw API cost)Budget 25% above raw API cost for realistic planning.
FAQ
How much does a 10-second video cost?
$0.20 (Kling Turbo Standard, 720p) to $1.00 (Kling Omni Pro, 4K). MiniMax H3 is in the $0.30-$0.80 range but includes audio. Pick based on quality needs and whether you need synchronized audio.
Is 4K worth the cost premium?
For social media: no. Instagram/TikTok/Douyin compress to 1080p max. You’re paying 2.5× more for pixels that get discarded. For website hero videos or presentations: possibly worth it.
How do I estimate cost for variable-length videos?
With Kling’s per-second model: multiply duration × rate. With H3’s flat-rate model: cost is per-generation regardless of length (within tier limits). For budgeting, assume your average duration and multiply.
Can I generate video on SandBase?
Yes. MiniMax H3 (text/image/reference-to-video), Kling Video 3.0 (all tiers), and Gemini Omni Flash are available through the SandBase ecosystem via /v1/run. The pricing in this article reflects provider rates; check the model catalog for current SandBase rates.
What’s the quality difference between $0.10 and $0.70 per video?
Significant. Turbo Standard (720p, $0.02/sec) produces recognizably AI-generated video: slightly soft motion, occasional artifacts, limited detail. Omni Pro ($0.07/sec, 1080p+) is close to stock footage quality: smooth motion, consistent lighting, natural physics. Test both on your actual use case.
For the broader pricing landscape including LLMs and data APIs, see our LLM pricing guide and social media data API guide.
Key takeaways
- Video generation pricing varies 10× across tiers: $0.10 to $1.00 for the same 10-second video
- Per-second pricing (Kling) is most predictable; token-based (Gemini) is least predictable
- MiniMax H3’s included audio saves 10-15% vs doing video + audio separately
- Budget 25% above raw API cost for retries, prompt iteration, and storage
- For social media content: Turbo Standard (720p, $0.10/5s) is the sweet spot — platforms compress to 1080p anyway
- The quality-cost decision should be made with A/B test data, not assumptions
