Disclosure: some links on this page may be affiliate links; we may earn a commission at no extra cost to you. Read the disclosure.

By the toolist.ai editors. Researched from Luma AI's own pages and hands-on tests published by others, not hands-on tested by toolist.ai. Updated October 15, 2026.

Luma AI review: pricing, features and what testers found

Luma AI is Luma Labs' creative app, formerly Dream Machine, where agents and Luma's own Ray3.2 video and Uni-1 image models share one credit balance with Kling, Seedance, Veo 3.1 and MiniMax H3. Ray3.2 makes silent 5- or 10-second clips at up to 1080p with up to 16 keyframes, restyles existing footage and exports 16-bit HDR and EXR. Paid plans start at $30 a month ($25 a month billed yearly), and the Luma API sells Ray3.2 from $0.15 a clip.

Our take

Luma AI suits creators and agencies who want to direct silent shots precisely and hand them to post-production: its Ray3.2 model takes up to 16 keyframes per clip, restyles existing footage for up to 20 seconds at 1080p, and exports 16-bit HDR and EXR. Plus costs $30 a month ($25 billed yearly) for 10,000 credits, which by our calculation buy 500 seconds of Ray3.2 at 720p in 5-second clips (6 cents a second) or 125 seconds at 1080p, and the same credits also run Kling 3.0, Seedance, Veo 3.1 and MiniMax H3. Ray3.2 makes no sound of its own. Curious Refuge's Brian Dalton scored Ray 3.2 7.7 out of 10 in June 2026, a step up from Ray3.14 (7.51) but behind Seedance 2.0 and Kling 3.0, with physics-heavy shots breaking down. There is no free plan beyond trial credits, unused plan credits do not roll over, fees are non-refundable, the terms let Luma train its models on paid users' uploads and output, and neither Ray3.2 nor Ray3.14 is on Artificial Analysis's video boards yet.

Based on Luma AI's published plans and the hands-on tests linked below, not on a test of our own.

Good for

  • Agencies and brand teams making many variants of a campaign, with agents, shared boards and Skills that repeat a workflow
  • Filmmakers and VFX artists who want keyframe control, footage restyling and HDR or EXR plates for grading and compositing
  • Trying Ray against Kling, Seedance, Veo 3.1 and MiniMax H3 on one credit balance: Plus's 10,000 credits also buy 204 seconds of Kling 3.0 at 1080p with sound (49 credits a second) or 35 seconds of Veo 3.1 with sound (280 credits a second)

Watch out for

  • Ray3.2 and Ray3.14 make no sound: add voice, music and effects as a separate step, or switch to Seedance, Kling, Veo 3.1 or MiniMax H3, which make sound. At 720p (768p for MiniMax) all of them cost more credits a second than a 5-second Ray3.2 clip (20), but at 1080p Kling 3.0 with sound (49) costs less than silent Ray3.2 (80). A 10-second Ray3.2 clip costs three times a 5-second one, so 50% more per second.
  • Plan credits expire at the end of each billing cycle and do not roll over. Credits for failed generations come back, but outputs you reject are billed, and Luma's own guide warns that the Agent may make extra variations or retries unless told not to. On-Demand Spend carries on past the plan at a rate the pricing page does not publish.
  • No free plan, only trial credits of an unpublished amount. Fees are final and non-refundable unless Luma decides otherwise; subscriptions renew automatically and a cancellation must arrive at least a day before renewal.
  • Luma's terms (updated May 14, 2026) give it a licence to use paid subscribers' uploads and outputs to train its models, perpetual for outputs; for free use the licence widens to displaying and distributing them publicly. Luma's Enterprise page says nothing Enterprise customers make is used for training, and the API's Scale tier advertises a no-train guarantee.
  • Luma's pages disagree: the pricing page says yearly billing saves up to 20% but each plan saves about 17%; it lists a Team plan that Luma's notes for AI assistants say does not exist; the API page prices a 5-second 720p video-to-video edit at $1.44 in its table and $1.08 in its FAQ; and the iPhone app, still named Luma Dream Machine and last updated April 17, 2026, lists older Lite, Plus and Unlimited in-app purchases at $9.99, $29.99 and $94.99.
  • No blind-vote ranking for the current models: on October 15, 2026 neither Ray3.2 nor Ray3.14 was on Artificial Analysis's text-to-video or image-to-video boards; the older Ray 3 sat 47th of 82 on its silent image-to-video board (Elo 1215).

What Luma AI does

Price

Free plan: None; the pricing page lists no free plan, only free trial credits to start (Luma does not publish how many). Luma's terms allow commercial use only of output made on a paid subscription, and give Luma wider rights over free users' uploads and output, including to display and distribute them publicly.

PlanBilled monthlyBilled yearly, per monthIncludesPrice per second of Ray3.2 720p video
Plus $30 $25 10,000 credits a month: 500 seconds of Ray3.2 at 720p in 5-second clips (100 credits each), 330 seconds in 10-second clips (300 credits each) or 125 seconds at 1080p in 5-second clips (400 credits each); Luma's and other companies' image and video models, commercial use, edit access for guest collaborators; $300 billed yearly $0.060
Pro $90 $75 40,000 credits a month: 2,000 seconds of Ray3.2 at 720p or 500 at 1080p in 5-second clips; "4x usage" of Luma Agents; $900 billed yearly $0.045
Ultra $300 $250 150,000 credits a month: 7,500 seconds of Ray3.2 at 720p or 1,875 at 1080p in 5-second clips; "15x usage" of Luma Agents; $3,000 billed yearly $0.040
Team and Enterprise not listed not listed Contact sales. Team adds team members, projects, usage analytics, shared team credits and SSO; Enterprise adds enterprise commitments, training and custom fine-tuning –
Luma API not listed not listed Pay per Ray3.2 clip in SDR: text- or image-to-video $0.15 at 540p, $0.30 at 720p or $1.20 at 1080p for 5 seconds, and $0.45, $0.90 or $3.60 for 10 seconds; video-to-video from $0.72 for 5 seconds at 540p; Reframe $0.06 to $0.36 a second; HDR doubles the price and HDR with EXR triples it. Uni-1.1 images from $0.0404 at 2K. A Scale tier sells dedicated capacity with an SLA and a no-train guarantee –

Price per second of Ray3.2 720p video is our calculation: the monthly-billing price divided by what the plan includes, assuming you use all of it. It is not a measured cost.

Prices last checked: 2026-10-16 on Luma AI's pricing page. Full breakdown: Luma AI pricing.

Price history

No price change since we first recorded these prices on October 15, 2026. We re-check them every week and list each change here.

What hands-on testers found

We have not tested Luma AI ourselves. These are the published tests we found that used it hands-on, newest first; follow the links for the full reviews.

Curious Refuge · Brian Dalton · June 11, 2026

What they tested: Ray 3.2 on Curious Refuge's standard set of realistic, animated and effects prompts, scored for prompt adherence, temporal consistency, visual fidelity, motion quality, and style and cinematic realism.

  • Prompt adherence was its best category (8.1 of 10), with nine of seventeen prompts scoring 9, though it dropped small beats such as a character reaching out to touch a chain.
  • Slow, human movement held together, with landscapes and hero shots scoring 9, but physics-heavy shots broke down: an explosion looked flat, a hockey player was swallowed by the frame during a fast camera move, and in a 2D scene a cup hung in mid-air.
  • He judged it a real improvement on Ray 3.14 and recommended it to creators who work from shot lists, but placed it below Seedance 2.0 and Kling 3.0.

Their score: 7.7 out of 10

Curious Refuge · Brian Dalton · February 24, 2026

What they tested: Ray3.14 on many of the same prompts, scored in the same five categories.

  • Prompts written as concrete, filmable actions did far better than vague ones: an over-the-shoulder conversation scored a perfect 10 for adherence, a loosely worded 2D animation 7.
  • A clear glass filling with juice in bright light made the edges shift from frame to frame, and style and cinematic realism scored lowest (7.33).
  • He found Kling 3.0 a little smoother but Ray3.14 better at following blocking and contact physics, and cautioned that loosely worded prompts tend to come out with physics that falls apart.

Their score: 7.51 out of 10

What users say

Test result

toolist.ai has not run its own test of Luma AI; the findings above come from the published tests linked. See how we research and test.

Best text to video tools · Luma AI pricing

Do you make Luma AI? Claim this listing to correct it; it's free. To see whether ChatGPT, Claude, Gemini and Perplexity recommend it, and how many readers we send you: toolist.ai for AI companies.

Sources