- π Krea's 2026 image model, generation plus instruction-driven editing
- β‘ Krea states most edits return in under 8 seconds
- π Krea reports native 2K output via its image API
- π― Krea reports consistent products and characters across images
- π§ Diffusion transformer with VLM-based prompt expansion, per Krea
- π§ Selectable thinking modes and multi-reference image inputs
- π’ Built by Krea.ai, successor to the Krea 2 line
Krea is a company building generative artificial intelligence products, with a focus on image generation and creative visual tooling. Its work centers on producing high-quality AI imagery through models designed for both fidelity and speed.
Explore 3 more models by Krea βKrea 3 is the latest image model from Krea.ai, the generative creative-tools company behind the Krea 2 family. Released in 2026, it is a generation and instruction-editing model, and ships alongside the dedicated editing variant Krea 3 Edit. In its announcement, Krea says most edits complete in under eight seconds at native 2K, with products and characters staying consistent from one image to the next.
Compared with the earlier generation β Krea v2 Large, Krea v2 Medium and the speed-tuned Krea 2 Turbo β the main shift Krea describes is from style-led generation toward precise, prompt-specified edits on an image you supply, while preserving everything the instruction did not ask to change. Krea's developer documentation shows the API accepting reference images, a target resolution, and a selectable thinking mode, so callers can trade deliberation time against latency per request.
Technically, Krea's own report describes a diffusion transformer paired with an in-house post-trained vision-language model for prompt expansion, plus a DFlash2 speculator that Krea's technical report reports at 700β900 tokens per second. The same report describes sampling optimizations such as activation quantization, context parallelism, timestep caching and optimized classifier-free guidance, and training with FSDP sharding, gradient checkpointing and bfloat16 optimizer moments. Krea also notes that training instructions enumerate every intended change, so the model learns not to perform edits the prompt never requested.
This About section is AI-generated from public sources via VeniceStats + Venice inference, with no human editing. It may contain inaccuracies.
Daily usageΒ Β·Β Krea 3Β Β·Β Organic providers
| Provider | Health | Repβ | Sales | Images | Routing | $ / img | Save |
|---|---|---|---|---|---|---|---|
| Venice.ai Proxy 0x1f22β¦18c9 | healthy | 75.07 | $0.15 | 4 | #1 | $0.0165 | β76% |
"Best price" and the provider table are live Antseed catalog data (advertised $/1M tokens β or $ per generated image for unit-billed image models β not settled amounts). Reputation = buyer trust score (0-100, the Antseed SDK's own formula). "Routing" = the SDK's default buyer routing (what Antseed Desktop ships with): a trust β₯ 60 gate on the effective reputation, then cheapest-first among routable providers; live failover state (per-peer cooldowns) is buyer-side runtime and not included. Model knowledge (TLDR, provider, About) via the VeniceStats enrichment layer. Advertised catalog, not the model used in any specific purchase. "Usage on Antseed" counts only settlements whose buyers share the per-model split on-chain (metadata v2/v3, opt-in), so every usage figure is a lower bound.