Models
Every model behind Likna, compared
Likna doesn't make you pick a model soup — it names the outcome and picks the right engine underneath. Here's what's actually running: four still models and one video model render every request today. Wan 2.2 and Kling v3 Pro are two video engines Likna ran through August 2026 and has since retired — they keep pages too, so you can see what they were and why Likna moved on. Every model below covers speed, cost, and sample output.
Z-Image Turbo
A 6B local model tuned for fast, natural everyday shots — Likna's default for most stills, rendered entirely on Likna's own GPU. See sample output, pricing, and how it compares below.
FLUX.1 dev
The high-detail image model in Likna's lineup — the same FLUX.1 dev weights whether Likna renders on fal.ai's cloud (fal-ai/flux-lora) or on its own GPU. Samples below are plain FLUX.1 dev output; identity conditioning is a separate feature, on its own page. See sample output, pricing, and how it compares below.
NoobAI XL
The illustration model in Likna's lineup — an SDXL base trained on Danbooru/e621 tags, so it draws anime and illustrated characters natively instead of fighting a photoreal prior. Trained avatars run on it the same way they run on FLUX. See sample output, pricing, and how it compares below.
Face Lock
PuLID identity conditioning on the FLUX family — lock one reference face and it holds across every scene, outfit and lighting setup, instead of drifting into a different person shot to shot. See sample output, pricing, and how it compares below.
MiniMax H3
Likna's video engine — MiniMax H3, rendered on fal.ai's cloud, with native audio included at no extra charge. Pick one of two resolution rungs per generation: 768p or 2K, each priced per second. Open-weight and genuinely self-hostable: Likna also runs the same weights on its own hardware, though that leg isn't part of the paid video flow today. See sample output, pricing, and how it compares below.