Live worlds.
Branching stories.
For a viewer-driven narrative or interactive prototype, choose a persistent session you can steer while it plays.
Start with H3 Max DirectorCompare video costs and workflows across H3 Max, Veo, Kling and Wan.
Start by comparing finished clips. Ordinary Max and Turbo use the same basic request-and-file workflow as Veo. Director is an additional live-session option with its own minimum bill.
Ordinary H3 Max has no Director session floor. Compare output price and resolution before choosing a live workflow.
Scroll sideways to compare all models
| Finished clip API | H3 Max · fal 768p | H3 Max Turbo · fal 768p | Veo 3.1 Fast · 720p |
|---|---|---|---|
| Output price / second | $0.02 launch / $0.08 standard | $0.01 launch / $0.04 standard | $0.10 standard |
| One 6-second clip | $0.12 launch / $0.48 standard | $0.06 launch / $0.24 standard | $0.60 |
| 60-second session floor | No | No | No |
| Output and audio | Finished file · native audio | Finished file · native audio | Finished file · native audio |
| Standard-price date | Sep 15, 2026 | Sep 15, 2026 | Published standard rate |
| Provider source | fal H3 Max | fal Turbo | Google pricing |
768p and 720p are different resolutions. These are price comparisons, not an independent quality benchmark. Max / Turbo dates follow fal’s overview; reference-to-video uses another billing formula.
Scroll sideways to compare all four models
| What matters | REALTIMEH3 Max Directorfal · 768p session | Veo 3.1 FastGemini API · 720p | Kling 2.6 Profal · with audio | Wan 2.6Alibaba Cloud · 720p¹ |
|---|---|---|---|---|
| Price per second | $0.02Launch → $0.08 standard | $0.10720p with audio | $0.14$0.07 with audio off | $0.10US endpoint, with audio |
| Delivery & speed | Live WebRTC stream | Async video file | Async video file | Async video file |
| Change prompts live | Yes, during a session | New request or extension | New request | New request |
| Character / scene context | 12 prior segments by default; up to 50 | Reference images & video extension | Prompt and image-based controls² | Prompt / reference workflows² |
| Continuous playback | Ongoing session | Clips; extension available | Separate clips | Separate clips |
| Generated audio | Included · 48 kHz stereo | Included | Optional; priced above | Included in this endpoint |
| Minimum / duration | 60-second billing floor$1.20 launch / $4.80 standard | 4, 6 or 8-second clips | 5 or 10-second clips | Separate generation jobs |
| 900 seconds of output³ | $18 / $72One 15-minute session | $90Multiple requests | $126Multiple requests | $90Multiple requests |
| API integration | WebRTC client + server proxy | Gemini SDK + job polling | fal SDK + job polling | Async API + job polling |
| Published source | fal Director | Google pricing | fal Kling pricing | Alibaba pricing |
¹ Wan uses wan2.6-t2v-us, US (Virginia), 720p. Region, endpoint and audio settings affect prices. These are named configurations, not a resolution-matched quality benchmark.
² Reference or image-based controls can require a different endpoint in the same model family. None of these controls guarantees perfect consistency.
³ Clip-model totals are simple output budgets for 900 seconds across multiple requests. They do not describe one continuous session or include retries, edits, extension overhead or taxes. Rendering latency varies with provider load and settings.
Among the endpoints compared here, Director is the one built around an ongoing stream. That changes how you design an interactive app.
Veo can extend a clip, but extension is another generation operation. It does not provide Director’s live prompt channel.
Check Veo’s actual capabilitiesA 4-second Veo 3.1 Fast clip at $0.10/s is $0.40. A new Director session stopped within its first minute is still $1.20 at launch or $4.80 at standard price. For longer output, compare per-second rates; for quick experiments, compare actual bills.
Calculate your costFor a viewer-driven narrative or interactive prototype, choose a persistent session you can steer while it plays.
Start with H3 Max DirectorFor reference-guided clips and video extension, test Veo 3.1. Judge output on your own prompts before committing a production budget.
Test Veo 3.1 / FastCompare fal H3 Max clips, Wan and Kling by the exact resolution and audio setting. Kling’s silent rate and Wan Flash can change the math.
Compare clip endpointsA little clarity before you hit generate.
Start with a short brief that represents the work you will actually publish: a product shot, a speaking subject or a moving scene. Specify the required duration, delivery resolution, audio and reference inputs before comparing prices. Generate a small sample with each candidate and record both the amount spent and the number of clips you would keep. Dividing that spend by usable clips gives a more practical production estimate than comparing the cheapest advertised second alone. Keep the launch and standard rates in separate columns when planning repeat work. For an interactive project, also evaluate how the live session fits your application and how often users are likely to start a new session, because each Director session has its own billing floor. For a finished video assembled from clips, allow for editing and continuity checks between shots. This evaluation method does not assume one model wins on quality; it makes your own acceptance criteria part of the cost comparison.
Ordinary H3 Max at 768p is $0.02/second at launch or $0.08 at standard price on fal, compared with $0.10 for Veo 3.1 Fast at 720p with audio. That is 80% or 20% lower per output second, at different resolutions. Turbo has separate lower rates. Director shares the quoted session rate but adds a 60-second minimum, so its short-session bill is different.
Veo 3.1 supports reference images and video extension. Those controls can help maintain a subject across generated clips. They are a different workflow from Director’s ongoing session context and live prompt changes; neither is a guarantee of perfect visual consistency.
Only the Director figures describe a single public streaming session. Other 15-minute figures are arithmetic budgets for 900 seconds of generated output across multiple requests. They do not promise a single 15-minute clip, seamless continuity or a fixed rendering time.
Provider pages checked Sep 9, 2026. Every comparison names the endpoint and resolution. Calculated totals are estimates for generated output, excluding taxes, retries and other services. We do not run an independent video-quality benchmark.