Makernice
← All posts
Four glass hourglasses of different heights on a slate shelf, sand running in each at a different pace, cool morning lightComparisons

What Are the Best Crun Alternatives for Video APIs in 2026?

Makernice Team · ·

Key takeaways

  • reAPI, WaveSpeedAI, Makernice and CometAPI each carry most of the Seedance, Kling, Wan and Happy Horse models on Crun, and none of them needs a subscription.
  • Crun's credit rate per dollar rises with the size of the top-up plan, its listed costs are estimates settled on actual usage, and it accepts up to 20 new generation requests per 10 seconds.
  • reAPI allows 100 concurrent media tasks per account by default, WaveSpeedAI raises limits by account level, and Makernice runs 2 generations at once by default with a price fixed before each one.
  • Stay on Crun if its free trial, its 14-day file window and a single asynchronous task flow already cover your volume.
Ask an AI assistant about this pageAsk ChatGPTAsk ClaudeAsk Perplexity

For the Seedance, Kling, Wan and Happy Horse models you run on Crun, the strongest alternatives are reAPI, WaveSpeedAI, Makernice and CometAPI. All four are pay-as-you-go and list most of the same model versions. Moving makes sense when Crun's estimate-first pricing, its submission ceiling or its two-week storage starts to bind. The pages behind every statement here were checked on November 11, 2026.

Where Crun's credit model gets in the way

Crun sells credits in four top-up plans, and the bigger the plan, the more credits each dollar buys: 200 on the entry plan, 224 on the largest. That makes the real rate of a model depend on how you topped up. Its rate cards also say "The following are estimated credits and costs. Final charges are based on actual official Token usage." So the number on the page is a guide, and the charge settles after the run.

Operational limits come next. All generation tasks on Crun are asynchronous, the documented ceiling is "Up to 20 new generation requests per 10 seconds", and generated media files are kept 14 days before automatic deletion. Reference-video modes on Seedance are priced per second of input plus output. We found no rule on failed tasks or cancellation, and consumed credits are non-refundable.

Which pay-as-you-go APIs carry the models you run on Crun?

reAPI, WaveSpeedAI, Makernice and CometAPI are pay-as-you-go, multi-model APIs that carry most of the Seedance, Kling, Wan and Happy Horse versions on Crun, with no subscription on any of them. reAPI and WaveSpeedAI suit high task volume, Makernice suits jobs that need a set price, and CometAPI suits accounts that also buy text models.

PlatformBest forHow it billsSource checked
reAPIBatch jobs that need many media tasks running at onceCredits at a fixed US-dollar rate, per completed generation; failed runs refundedreapi.ai · 2026-11-11
WaveSpeedAIGrowing API traffic where limits should scale with spendPay-per-use balance; estimate on screen, final task charge prevailswavespeed.ai · 2026-11-11
MakerniceJobs where the price must be agreed before submissionPrepaid US-dollar balance per output second or per image, one rate for every modeMakernice pricing · live price list
CometAPIAccounts that pay for text models from the same creditsPer token for text, per image or per second for media; no minimum spendcometapi.com · 2026-11-11

Per-second rates for each shared model and resolution, including Crun's, are worked out from the published lists on the Crun alternative page. Here we compare how the bill is formed.

What each alternative changes for a Crun user

reAPI keeps one credit rate for everyone, sold at a fixed US-dollar value, and bills each completed generation per token, image or video second. Failed generations are refunded in full, which closes the question Crun leaves open. Default account limits include 100 concurrent media tasks, credits last while the account is active, and new accounts start with free credits. Results come from polling a task endpoint, and assets sit in storage for a limited window set in its privacy policy.

WaveSpeedAI bills from a pay-per-use balance that never expires, refunds failed requests automatically, and shows an estimate on the Generate button with the note that "the final task charge prevails". Its rate limits climb through Silver, Gold and Ultra account levels, set by the size of a single top-up. There is no batch endpoint; concurrent single requests are tracked by polling or webhook, and files generally expire within 7 days, a shorter window than Crun's.

Makernice removes the estimate step for its 19 video and image models. A REST quote is locked for 12 minutes and the job is created from that quote; over MCP, generate prices the job when called and refuses it if the price is over your max_cost_usd. Video bills by output seconds at one rate whatever the mode, so a reference image does not add a separate input charge; its Seedance models take no video input. An account runs 2 jobs at once by default with no batch endpoint, failed and timed-out runs are not charged while an output blocked by moderation is, and only a queued job can be cancelled for a full release.

CometAPI suits Crun users who also buy text models, billing per token for text and per image or second for media from one credit pool with no monthly fee or minimum spend. A model error is not billed. Its model pages give test estimates for 5, 10 and 15 second clips. CometAPI states that it does not store model outputs, so results must be saved on arrival, and purchased credits are generally non-refundable.

Throughput and storage limits after you switch

Throughput after Crun depends on the platform: reAPI allows the most work in flight by default, WaveSpeedAI raises limits as account level grows, and Makernice runs 2 jobs at once per account. Storage gets shorter on WaveSpeedAI and longer on Makernice than the 14 days Crun keeps files.

  • Submission rate: Crun caps new requests at 20 per 10 seconds. reAPI's defaults are 50 requests per second and 100 concurrent media tasks; WaveSpeedAI ties limits to account level; Makernice runs 2 concurrent jobs per account by default.
  • File storage: Crun deletes media after 14 days and WaveSpeedAI generally within 7. Makernice outputs carry an expiresAt, 30 days after completion in the docs example, and CometAPI keeps no copy.
  • Getting results: Crun and reAPI are polling-based; WaveSpeedAI adds webhooks; Makernice offers polling, a per-job callback_url or an account webhook.
  • Agents: Makernice connects to Claude Code and Cursor as an MCP server; WaveSpeedAI links CLI and MCP tools from its footer.

When does staying on Crun make sense?

Crun is a fair choice when volume stays under its submission ceiling and two weeks of storage is enough. Its free trial lets a new user test before paying, and the larger top-up plans reward teams that commit a bigger sum up front. It also lists Qwen image models and Kling 3.0 4K next to Seedance in one async flow. If you are weighing a move model by model, the Crun alternative page shows the overlap.

Frequently asked questions

Why does Crun call its prices estimates?
Crun's rate cards say "The following are estimated credits and costs. Final charges are based on actual official Token usage." The charge is settled after the run, and the per-dollar credit rate also depends on which top-up plan you bought.
Which Crun alternative handles the most concurrent jobs?
reAPI's default limits allow 100 concurrent media tasks per account. WaveSpeedAI raises limits with account level, and Makernice runs 2 jobs at once per account by default.
Is a reference video billed differently on Crun and the alternatives?
Crun prices Seedance reference-video modes per second of input plus output. Makernice bills by output seconds at one rate across modes, and its Seedance models do not take video input.
Does Crun charge for failed generations?
We did not find a failed-task rule on Crun's pages. reAPI refunds failed generations in full, WaveSpeedAI refunds failed requests automatically, and Makernice does not charge failed or timed-out runs, though it does charge an output blocked by moderation.