← Image

KreaKrea

Krea

免费+订阅Free + subscription 实时生成Real-time generation 模型聚合Model aggregation LoRA训练LoRA training 自研线In-house model line partial lifecycle: active 核验Checked 2026-10-22

Krea 是「实时生成 + 多模型聚合」的创作台：边画边出的实时线是招牌，官网聚合 64+ 家模型（Veo/Ideogram/Flux 等）+ 自研 Krea 2 线；「Do not train」是官方营销口径。

Krea is the real-time + multi-model aggregation console: draw-as-you-see realtime line is the signature, 64+ third-party models aggregated (Veo/Ideogram/Flux…) plus self-developed Krea 2; 'Do not train' is marketing copy, terms govern.

适合与不适合When to use it, when not to

适合：实时交互创作、一平台切换多模型、云端 LoRA 训练。不适合：开放权重离线、单模型深度调参、把聚合当「全都最强」。

Fits: real-time interactive creation, switching among many models in one platform, cloud LoRA training without local setup. Not a fit: open weights, offline use, treating every aggregated model as best-in-class.

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-22

出品属地Origin: 国外出品foreign-made　Krea（伦敦）。Krea (London).
国内使用Direct access: 直连不可达direct blocked　krea.ai 大陆直连实测 000（2026-10-22，代理下 200 可达）——直连不可达，需网络环境。krea.ai measured 000 direct from mainland (2026-10-22; 200 via proxy) — unreachable without a network path.
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-22 检索）；该产品未面向大陆公众提供服务，备案不适用。not in the CAC public registry (searched 2026-10-22); the product does not serve the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 多模型聚合+自研混合：官网列 64+ 模型（Veo/Ideogram/Runway/Luma/Flux/Gemini 等，官网导航所见）；自研 Krea 2 线（官网命名）；实时生成线（官网文案「real-time as you type or draw」）。
Multi-model aggregation + in-house hybrid: the official site lists 64+ models (Veo/Ideogram/Runway/Luma/Flux/Gemini etc., as seen in the site navigation); the in-house Krea 2 line (named on the official site); a real-time generation line (site copy: "real-time as you type or draw").
运行位置Where it runs: 网页（krea.ai）；生成在厂商云端。未见官方离线权重线；实时线对延迟敏感，官方文案强调低延迟（具体 SLA 未核验）。
Web (krea.ai); generation on the vendor's cloud. No official offline-weights line seen; the real-time line is latency-sensitive, official copy stresses low latency (no concrete SLA verified).
本地文件Local files: 作品存账户侧云端；未见官方声明的本机数据面。训练产物（LoRA）能否导出到本地，官方页未见明确说法（未核验）。
Works stored account-side in the cloud; no officially stated local data surface. Whether training artifacts (LoRA) can be exported to local — no clear statement on official pages (unverified).
关机后的任务Tasks after shutdown: 实时生成是卖点（边输入/边画边出图，官网文案）；无「关机续跑」任务面（未核验到声明）。
Real-time generation is the selling point (output as you type/draw, official copy); no "keep-running-after-shutdown" task surface (no claim verified).
工具与扩展Tools & extensions: 实时画布、LoRA 微调（官网文案）、放大线（官网文案提及 22K 上限，抓取快照未见该字样——未逐条核验）、视频模型聚合（Veo/Runway 等经聚合面提供）；API 面存在，细节未逐条核验。
Real-time canvas, LoRA fine-tuning (official copy), upscaling line (official copy mentions a 22K ceiling; the fetched snapshot did not show that figure — not verified item by item), video-model aggregation (Veo/Runway etc. served through the aggregation surface); an API surface exists, details not verified item by item.
上下文与记忆Context & memory: 以项目/画板组织（官网导航所见）；跨会话记忆未核验到官方声明。实时画布的「当前画面」是会话内状态，不是持久记忆。
Organized by projects/boards (as seen in site navigation); cross-session memory — no official statement verified. The real-time canvas's "current frame" is in-session state, not persistent memory.
权限与限制Permissions & limits: 账号+订阅体系（D9：价格不采）；训练用途：官网营销文案主张不训练用户内容（「Do not train」），隐私页原文口径未逐条核验——营销话术与条款以原文为准。
Account + subscription system (D9: prices not collected); training use: official marketing copy claims user content is not used for training ("Do not train"); the privacy-page original wording not verified line by line — marketing talk vs. terms, the original text governs.
适合什么任务Best for: 适合：实时交互创作（边画边看）、一个平台切换多家模型、LoRA 训练免本地环境。不适合：要开放权重、要离线、要单模型深度参数控制。
Fits: real-time interactive creation (draw-and-see), switching between multiple vendors' models on one platform, LoRA training without a local environment. Does not fit: wanting open weights, offline use, or deep per-model parameter control.

头号误解Top misconception

Krea 是「聚合+自研」混合——「这个功能背后是谁的模型」要按官网模型清单逐项看，别按单一模型理解全站Krea is an aggregation + self-developed mix — 'whose model is behind this feature' must be read off the official model list item by item
实时线与前沿线是不同模型——实时档的速度换质量，别把两者质量预期混在一起The realtime line and frontier lines are different models — realtime trades quality for speed; don't mix quality expectations
「Do not train」是营销文案——训练与数据用途最终以隐私页原文条款为准（本站未逐条核验）'Do not train' is marketing copy — training and data usage follow the privacy/terms text (not verified line by line)

未知项清单Known unknowns

放大上限（官网文案提及 22K，抓取快照未见——未逐条核验）
训练条款原文与「Do not train」的条款级对应（未逐条核验）
API 的模型覆盖与配额（未逐条核验）
跨会话记忆机制（未核验到官方声明）

Upscaling ceiling (official copy mentions 22K, the fetched snapshot did not show it — not verified item by item)
The training-terms original text and its clause-level correspondence to "Do not train" (not verified line by line)
The API's model coverage and quotas (not verified item by item)
Cross-session memory mechanism (no official statement verified)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-22，内容更新日content updated on 2026-10-22。

类型Type	名称Name	链接Link
official	Krea · 官网（产品与能力叙事、模型清单）Krea · official site (product and capability narrative, model list)	https://www.krea.ai
changelog	Krea · Changelog（产品更新线）Krea · Changelog (product update line)	https://www.krea.ai/changelog
official	Krea · Privacy（官方隐私页，2026-10-22 直抓核验）Krea · Privacy (official privacy page, direct-fetched and verified 2026-10-22)	https://www.krea.ai/privacy

实测记录Measured runs

本档未跑生成质量与速度测试；结论只覆盖能力面与许可面。

实测建议（三条，都指向官方可读页没给的东西）：

实时线与前沿线的质量-延迟实际曲线（同一提示词对比）
聚合模型的输出是否被 Krea 后处理（放大/滤镜链）
LoRA 训练产物的导出与复用面

This site has not completed measured runs. The coding-oriented test protocol is at protocols/coding/_protocol.md (the coding task suite does not apply to creative products).

Suggested measurements (three, all pointing at things the official readable pages do not give):

The real quality-latency curves of the real-time line vs. the frontier line (same-prompt comparison)
Whether aggregated models' outputs are post-processed by Krea (upscaling/filter chains)
The export and reuse surface of LoRA training artifacts