Luma Uni-1.1
Image Available Full comparison ↗ luma/uni-1 · by Luma AI
· Autoregressive
Pricing — 1 offering(s)
About this offering: Luma markets the current version as Uni-1.1 (announced 2026-05-05), but the API's model id is still uni-1. The prices shown are Uni-1.1's.
Uni-1.1 · 2048px
- $0.0404 / image
Current
2026-07-23 → present
Text-to-image, no reference images. effective_from is the date this entry was sourced, not a confirmed…
Text-to-image, no reference images. effective_from is the date this entry was sourced, not a confirmed price-change date. Exact match to Krea's own $0.0404 listing (the source that surfaced this gap) — a rare case where the third-party figure and Luma's own page agree precisely. Image edits and reference-guided generation cost more on the same page: $0.0434 (edit or 1 reference image) up to $0.0644 (8 reference images), not separately tiered here.
Uni-1.1 Max · 2048px
- $0.10 / image
Current
2026-07-23 → present
Text-to-image, no reference images. Uni-1.1 Max's own reference/edit pricing on the same page ranges…
Text-to-image, no reference images. Uni-1.1 Max's own reference/edit pricing on the same page ranges $0.1030-$0.1240, not separately tiered here.
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Luma Uni-1.1 fits into a cost-aware routing setup
See how →Capability profile
How Luma Uni-1.1 rates across core capability dimensions, with the task-level evidence behind each rating.
Core marketed differentiator — Luma's own materials claim the unified reasoning+generation architecture 'interprets complex, multi-part prompts more accurately' than models that treat understanding and generation as separate stages. Vendor-stated; no independent benchmark cited here.
Luma's own site claims 'intelligent scene completion and spatial reasoning' — vendor-stated.
~31 seconds per image per independent reporting — slower than typical diffusion models, consistent with sequential autoregressive token generation rather than parallel denoising steps.
2048px (2K) output per Luma's own pricing page — no higher-resolution tier found.
Ratings are estimated — limited independent data is available for this model.
Operator guidance
Route here when prompt fidelity on complex, multi-part instructions is the priority — Luma's own materials position the unified reasoning+ generation architecture specifically against models that "simply match keywords." The Max tier trades roughly 2.5x the cost for higher quality. Slower per-image (~31s) than typical diffusion competitors, so not the fit for high-volume, latency-sensitive generation — Google's Nano Banana 2.1 or a distilled FLUX/SD tier are better suited there. Luma Photon, the predecessor image line, is deprecated per Luma's own guidance and not tracked in this registry.
Use cases
- Complex, multi-part, or spatially-precise prompts where literal instruction-following matters more than raw aesthetic polish
- Reference-guided generation and natural-language image editing in one API (up to 9 reference images per request, per Luma's own docs)
- Workflows already using Luma's Ray video API wanting one vendor for both image and video generation
Limitations
- "Reasoning" is Luma's own marketing/architectural framing (confirmed on Luma's own site, not fabricated by aggregators) but no independent technical paper confirms an explicit reasoning step distinct from autoregressive token prediction itself
- Slower generation (~31s/image per independent reporting) than diffusion-based competitors, a real tradeoff of the sequential token-prediction approach
- Luma's own documentation shows internal drift — one page (llm-info) explicitly deprecates the prior Photon models in favor of Uni-1.1, while another (docs.lumalabs.ai/docs/image-generation) still lists photon-1/photon-flash-1 as the default request examples