Short answer: Pick Pika H3 Flex when the job is multi-reference / multi-shot iteration — up to 12 image/video/audio refs, sync stereo, optional five audio stems, open weights for local runs, or a per-output-second API (480p $0.01/s … 4k $0.08/s). Flex is entering the Pika creation product on November 2 (announced Oct 29, 2026). Pick LongCat’s pay-once continuation pack when you already have a shot and need a coherent extend without wiring an API key or a GPU stack: Try $9.9 / 500 credits covers one LongCat 720p Standard job at 300 credits with 200 left. This is not another Kling-vs-continuation rewrite — see siblings 720p pack math and continuation vs Kling 4.0 30s if that is your question.
Three lanes (don’t mix the meters)
| Lane | What you get | How you pay | Best when |
|---|---|---|---|
| OSS / local weights | Flex generator adapters + inference (T2V, first/last frame, multimodal ref-to-video with sync audio at 480p/768p). Hosted-only extras stay hosted: 1080p+ upsampler, stem separation, prompt enhance, accelerated stack. | Your GPU, power, ops — not LongCat credits | You already run local inference and want to customize under the model license |
| API billed $/second | Hosted Flex image-to-video (and related Flex surfaces): duration 5–15s; only successful gens charged | 480p $0.01/s · 768p $0.02/s · 1080p $0.035/s · 2k $0.04/s · 4k $0.08/s | You need 12-ref / multi-shot / stem workflows and are fine metering dollars per second |
| LongCat pay-once continuation pack | Hosted text-to-video, image-to-video, and continuation (extend a coherent shot). Credits forever on one-time packs (except the 12-mo image pass) | Try $9.9 / 500; Pro $199.90 / 12,000; Max $389 / 50,000; Studio $499.90 / 64,000 (pack cards 2026-11-01) | You want one more beat on an existing clip without API keys or self-host |
LongCat’s page also lists Kling v3 as seconds × 1000 / 2000 (audio × 1.5). Do not fold those multipliers into LongCat continuation’s 300 cr math — same rule as the Kling sibling post.
Hosted LongCat generation needs credits; OSS self-host of LongCat weights is a separate path from the credit packs.
When you need up to 12 refs, multi-shot, or audio stems
Flex is built for iteration with receipts: character sheet + world still + motion clip + voice sample in one generate (up to 12 reference files spanning image, video, and audio). Multi-shot prompting, first/last-frame control, and reference-guided edits are the product shape — not “append the next LongCat segment.”
Audio: Flex can emit sync stereo with the video. Through the Pika API, you can request five stems — Dialogue, Score, SFX, Foley, Ambience — so a mix can change without regenerating the picture. Stem export is a hosted path (with the catalog’s separation handling); local open weights cover sync audio at native 480p/768p, not the full hosted finishing stack.
Resolution: Native generate at 480p or 768p; finishing at 1080p / 2K / 4K uses Pika’s hosted upsampler. If your brief is “draft at 768p, finish one hero at 4K, keep Dialogue separate from Score,” Flex (API or product surface after Nov 2) is the lane.
If your brief is only “keep this character lit the same for the next beat,” you do not need twelve refs — you need continuation.
When Try $9.9 / 500 credits continuation is enough
From longcat-video.org/pricing pack cards (2026-11-01 Asia/Shanghai):
| LongCat Video tier | Credits / job |
|---|---|
| 480p Distilled / Standard | 150 / 200 |
| 720p Distilled | 250 |
| 720p Standard | 300 |
| Image / image-edit row | 80 |
Try $9.9 / 500 credits (forever, one-time):
- One 720p Standard = 300 cr → 200 cr left (enough for an image row, or a Distilled continue — not a second Standard 720p).
- Or lean Distilled 720p @ 250 if you are still validating subject lock.
- Title framing on the site: from $9.90 once or $29.90/month — use the one-time Try math for a first continuation test.
Enough when:
- You already like the opener and need one coherent extend.
- You do not need 12-file multimodal control or stem stems today.
- You want a hosted answer without standing up Flex weights or an API org balance.
Step up when leftover 200 cr is the annoying part: Pro $199.90 / 12,000, Max $389 / 50,000, or Studio $499.90 / 64,000 on the same cards. For pack-depth math, stay with the 720p pack sibling.
Worked API cost sketch (official $/s × duration × resolution)
Only successful generations are charged. Duration window on the Flex image-to-video docs: 5–15 seconds. Illustrative stickers (no CTR, no invented volume):
| Output | Rate | 5s | 10s | 15s |
|---|---|---|---|---|
| 480p | $0.01/s | $0.05 | $0.10 | $0.15 |
| 768p | $0.02/s | $0.10 | $0.20 | $0.30 |
| 1080p | $0.035/s | $0.175 | $0.35 | $0.525 |
| 2k | $0.04/s | $0.20 | $0.40 | $0.60 |
| 4k | $0.08/s | $0.40 | $0.80 | $1.20 |
Reading the table against LongCat Try: a handful of successful 768p × 10s Flex API clips (~$0.20 each) can stay under $9.9 in raw generate dollars — but that meter buys Flex generation, not LongCat continuation subject lock. Ten failed LongCat continues still return credits on failed jobs per LongCat’s FAQ; Flex charges successful output seconds only. Compare job shape, not a fake “same dollar = same shot.”
Speed claim care: Pika’s “faster than real time” footnote is transformer (DiT) compute for a 15s @ 768p clip ≈ 7s in a controlled benchmark — not end-to-end API wall-clock. Refs, upsampler, stem separation, and service overhead change completion time.
UNVERIFIED soft bits: docs note stem output_mode can add a separation fee and bill decoded stem durations — exact additive sticker not treated as a fixed public rate card here. Prefer the $/s table above for planning.
FAQ
Does LongCat Try $9.9 fund a 720p Standard continue?
Yes this window: 720p Standard = 300 credits, so Try $9.9 / 500 funds one job with 200 credits left.
Is Pika H3 Flex the same job as continuation?
No. Flex is multi-reference / open-weights / $/s API generation. Continuation extends a coherent shot you already have.
Should I use LongCat’s Kling multipliers for continuation math?
No. LongCat Kling v3 rows use seconds × 1000 / 2000 (audio × 1.5). Do not fold those into LongCat continuation’s 300 cr math.
When do I pick Flex API instead of a LongCat pack?
When you need up to 12 refs, stems, or open-weights control and are fine metering $/output-second. Prefer LongCat packs for hosted continuation without API keys.
CTA
If the wall is continue the shot you already have, buy Try — $9.9 / 500 credits, run one 720p Standard (300 cr) continuation, and keep the 200 for a Distilled pass or an 80 cr image — then upgrade only when leftover credits, not model quality, are the bottleneck. If the wall is 12 refs + stems + open weights or $/s API, use Flex’s lanes; come back to LongCat when the next problem is coherent extend, not multimodal cold-start.
Sources
- https://pika.art/blog/pika-h3-flex — Oct 29, 2026 launch post; Nov 2 creation-product note; 12 refs; stems; open weights; DiT footnote
- https://dev.pika.art/llms/pika/pika-h3-flex/image-to-video — model id
pika/pika-h3-flex/image-to-video; $/s table; 5–15s; successful gens only - https://github.com/Pika-Labs/Pika-H3-Flex — open weights / inference structure cite
- https://longcat-video.org/pricing — Try / Pro / Max / Studio / Unlimited Images cards; LongCat credit tiers; Kling v3 multipliers listed separately
UNVERIFIED / soft
- FAQ text that still describes Max as 30,000 credits/month or 360,000/year (or older Pro $89 / 5k narratives in prior drafts) — prefer visible pack cards: Try $9.9/500, Pro $199.90/12,000, Max $389/50,000, Studio $499.90/64,000
- Exact additive audio-stems separation fee beyond the published $/s rates
- End-to-end Flex API latency (only DiT bench is published)
- Search volume / CTR: unavailable — not invented
Publish: no. Structure-only vs competitors; no rewrite of Kling-continuation or prior LIVE batches.