Ricebowl

Ricebowl AI does not train or run its own generation models. Its own privacy policy states generation requests are sent to "AI infrastructure partner kie.ai," which routes them to the underlying provider (Google for Veo, ByteDance for Seedance, Kuaishou for Kling, and others for the image models). Ricebowl's product is the credit wallet, the single login, and the workspace UI that sits in front of kie.ai's routing. This matters for reliability and pricing: Ricebowl's uptime and model access depend on kie.ai staying up and keeping its provider contracts, a layer buyers can't see into directly. kie.ai itself advertises 99.9% uptime and prices up to 88% below official provider APIs, which is the margin Ricebowl's own markup sits inside — kie.ai claims this, Ricebowl does not independently state its own margin (vendor claim, unverified by us).

Four monthly plans, verified on the pricing page 2026-10-03: Lite $12.99/mo (400 credits), Starter $29.9/mo (1,000 credits, 2 parallel tasks), Standard $49.9/mo (2,000 credits, 3 parallel tasks, marked 'Popular'), Advanced $89.9/mo (5,000 credits, 5 parallel tasks). Every plan unlocks all 39 models — there's no tier-gating by model, only by credit budget, and the page backs that up with a full table showing exactly how many renders of each specific model (20 credits for Kling 4.0 or MiniMax H3 up to 315 for Seedance 2.5; 3 credits for GPT Image 2.5 up to 14 for FLUX.2 Flex) each plan buys. That level of per-SKU disclosure is rare among credit-based AI tools, most of which quote a vague 'credits per month' with no table linking spend to output. The page advertises 'Save 30% billed annually,' but the annual price table loads behind a client-side toggle this run couldn't render, so take that 30% figure as the vendor's claim rather than a confirmed number. Failed generations refund their credits automatically, and the first subscription carries a 7-day money-back guarantee (refund policy, effective 2026-08-21).

OpenAI shut down the Sora app on 26 April 2026 and discontinued the Sora 2 API entirely on 24 September 2026 (confirmed via OpenAI's own help-center article and corroborated by Wikipedia and multiple trade press pieces). Ricebowl still lists "Sora 2" in its model picker, but both the homepage card and the dedicated model page disclose plainly that the entry auto-switches every request to MiniMax H3 and that no credits move until the user accepts that substitution. This is the kind of disclosure the house style here asks us to credit: Ricebowl is capturing residual search traffic for a dead model name without pretending to still offer it.

ricebowl.ai's own /about-us page names no founder, team, or company, and does not state when the product launched or where it is based. Its Terms of Service specify the agreement is "governed by and construed in accordance with the laws of the Hong Kong Special Administrative Region," with disputes arbitrated in Hong Kong — the only jurisdictional signal on the entire site. Searches for independent reviews (Trustpilot, Reddit) turned up nothing specific to ricebowl.ai as of 2026-10-03; this is a young, low-visibility product rather than one with negative signal to report.

The genuine strength is breadth plus honesty about cost: one balance spends across 39 current-generation video and image models (Veo 3.1, Kling 3/4, Seedance 1.5/2.0/2.5, MiniMax H3, GPT Image 2/2.5, Nano Banana, FLUX.2, Qwen Image, and more) without separate subscriptions or tier-locking any of them, and the pricing page shows exactly what each model costs in credits rather than hiding behind a vague monthly allowance. Model-specific landing pages (e.g. for Kling 3, Seedance 2, MiniMax H3) go further than most aggregators, documenting real technical trade-offs per model — reference-image limits, audio handling, resolution caps — rather than one generic 'AI video generator' pitch repeated on every page.