Text and reference-led images
Describe a subject, style and composition, or use an image you have permission to upload. Select a model and inspect its available controls.
O01O02Leonardo.Ai
Start with an idea, explore visual directions, then refine the images your game needs.
Evidence statusPublic-evidence reviewMedium confidenceLast checked Sep 7, 2026By MakeGameWithAI
PRODUCT OVERVIEW
What Leonardo.Ai is—and how it works.
Leonardo.Ai is a cloud-based visual creation platform. In a browser, you can turn a description or reference image into character concepts, props, backgrounds or illustrations, then edit, enlarge and download selected results. You do not need to draw the first version yourself or deploy an image model to get started.
It brings its own models and third-party models into one workspace. Image Creation offers model and generation settings; Flow State focuses on exploring many directions and making variations of an image you like. Reference controls, editing options and costs depend on the chosen model, so these are different routes rather than one uniform generator.
For game development, its clearest role is producing static-art candidates and visual references. Downloaded images can move into an art editor or your project, while an official API can automate generation and retrieval. Preparing transparent edges, sprite frames, engine settings and a consistent finished asset set remains separate work. The wider platform also offers other media; this review focuses on 2D images.
HOW YOU USE IT
Use the browser workspace with an account and available tokens. For automation, configure a Production API key and separate API balance; REST, official SDK methods or the documented MCP can be wrapped into an external agent workflow.
Flow State exploration
Explore candidates and use More Like This to pursue a direction. New results consume tokens even when you do not save them.
O05Editing and optional training
Refine selected images with supported edits or upscaling. Paid personal-model training is an advanced option, not a prerequisite for your first image.
O01O04O06Library and downloads
Keep useful results and download image files. Personal trained models are not downloadable, so asset portability and model portability differ.
O02O04Official API and MCP
REST covers generation, status and result URLs. The official MCP documents generate-image for Lucid Origin, Lucid Realism and Ideogram 3.0—not every Web feature.
O16O17O25INTERFACE & EXAMPLES
See how it works
Leonardo's official quickstart shows the prompt entry, creation tools, and featured Blueprints on the home screen.
Official source retrieved September 2026; interface version not specified.Source: Leonardo.Ai Help Center
REVIEW SCOPE
Basic non-confidential 2D creation and external generation/retrieval. Full animation consistency, 3D/video quality, runtime generation and production reliability are outside this assessment.
- Beginner friendlinessBeginner friendly
- Generate and download candidates from text or reference images in a browser, without first learning node graphs or deploying models. Advanced control, budgeting and final selection still take practice.
- Agent integrationEasy to integrate
- Official REST endpoints cover generation, status and result retrieval for external agents. API billing is separate; the official MCP server covers only selected models, and client compatibility has not been tested here.
Scope, integration requirements and sourcesChecked
Beginner scope: A browser user's first non-confidential 2D concept, character or prop candidate set, without a professional drawing background.
Agent scope: An external agent submits a 2D image generation, waits for the task and retrieves the resulting files; not development of an entire game.
Confidence: learning Medium · Agent Medium
- Leonardo REST APIOfficial
- Leonardo remote MCPOfficial
- Official Python SDK — ImageOfficial
- Access
- Configure an account, Production API key and API balance. REST uses Bearer authentication; the documented remote MCP uses an API-Key header. Never place the key in public code or a game client.
- Billing
- API PAYG is separate from Web tokens and from the external agent's model fees. Model/task quotes require the API calculator; the advertised trial credit's eligibility and rights were not verified.
- Getting results
- The v1 generation flow returns a generation ID. Query its status until COMPLETE or FAILED, then retrieve generated_images URLs. Preserve the ID and use bounded waiting; an empty PENDING response must not cause repeated paid submissions.
- Limitations
- The MCP documents only generate-image with Lucid Origin, Lucid Realism and Ideogram 3.0. REST model versions differ; v1/v2 schemas are not interchangeable. Actual client calls, throughput, accepted-asset cost and trial/PAYG-only rights remain unverified.
These are independent editorial judgments, not an overall score or a claim of hands-on agent integration testing.
- Official browser generation workflow · Official · Learning evidence
- Altmon reference-led character iterations · Independent · Learning evidence
- iXBT browser and Flow State walkthrough · Independent · Learning evidence
- API Quick Start Guide · Official · Integration evidence
- API General FAQs — status and result URLs · Official · Integration evidence
- Official Python SDK — Image · Official · Integration evidence
- Official Leonardo.Ai MCP Server · Official · Integration evidence
- Leonardo.Ai Terms of Service · Official
- API PAYG Guide · Official · Integration evidence
KEY FINDINGS
What the evidence supports—and what it does not.
Editorial inferenceMedium confidence
You can begin before you can draw
A character-development thread and a hands-on article describe reference-led iteration and browser exploration. They support the starting workflow, while a paid user's settings complaints show that learning and budget checks remain part of it.
O02U03U05U06User reportsMedium confidence
Art can enter a game, with other tools and human work
One developer reports using Leonardo for game imagery; an older Polycount project used it for magazine-cover art in a prop pack. These are scoped project accounts, not proof Leonardo produced the entire game or asset pack alone.
U02U04Editorial inferenceMedium confidence
A character illustration is not an animation pipeline
A GameDev.net author describes an unsuccessful multi-frame attempt involving Leonardo and Copilot. Its year and per-tool results are unclear: it is a caution about the task, not a current Leonardo failure rate. A successful static iteration does not establish matching motion frames.
U01U03Editorial inferenceMedium confidence
Agent access has a documented result path
The official REST flow returns a generation ID, exposes task status and provides image URLs. That is enough to support a scoped integration judgment. The MCP covers fewer models; neither the API documents nor this review guarantee every client's reliability.
O16O17O18O25EDITORIAL VERDICT
Our take
Medium confidenceLeonardo is a useful starting route when you have a game-art idea but cannot yet draw or configure a local image workflow. You can see alternatives early, choose a direction and keep improving it. Treat the result as a candidate to inspect, not a finished asset by default. Its official programmatic entry also makes it relevant to agent-assisted development, provided you account for separate API spending and the chosen model's limits.
Conclusion scopeMedium-confidence, public-evidence judgment about starting and iterating on static images—not an asset-quality ranking, acceptance-rate estimate or tested client integration.
BETTER FIT
Worth auditioning when
- First-time creators turning a non-confidential idea into visible art candidates.
- Game makers exploring a character, prop or environment direction before polishing it.
- Creators willing to compare models, inspect settings and make selective edits.
- Developers who want documented generation and result retrieval for an external agent.
POORER FIT
Do not depend on it yet when
- Expecting one prompt to deliver a complete consistent sprite animation set.
- Requiring a fully offline workflow or downloadable personal trained models.
- Depending on guaranteed exclusive rights without checking the plan, privacy mode and input permissions.
- Expecting a Web subscription to include API use, or every Web model to be available through MCP.
WORKFLOW FIT
Where it fits in a production workflow.
Explore before commissioning final art
Generate a few directions for one prop or background, select a promising silhouette or palette and keep the reference.
GuardrailStart with original, non-confidential prompts. Limit new generations, not just downloads.
Refine a selected candidate
Use supported reference/edit controls, then inspect the downloaded file at the size it will occupy in your game.
GuardrailBudget editing, cleanup and upscaling separately; verify text, edges and repeated character details.
Let an agent handle repeatable file work
Wrap the official API to submit a bounded request, retain its ID, await completion and save the returned image into a staging folder.
GuardrailKeep keys out of game builds; set spending limits and inspect results before promoting them into the project.
RECOMMENDED WORKFLOW
Use this tool inside a complete production workflow.
Start from a testable brief, then move through prototyping, assets, audio, testing, and release with an explicit handoff and human check at every step.
SHORTEST RESPONSIBLE PATH
Start with one prop or background direction.
This is a low-risk starting workflow synthesized from the evidence.
- 01
Choose one original, non-confidential visual idea; a finished animation set is not the first task.
- 02
Check available tokens, the active model, output count and public/private setting before submitting.
- 03
Generate a small candidate set, pick one direction and request a specific refinement.
- 04
Download useful results, inspect them at game size and record the settings and rights context.
- 05
Only then consider paid volume or API automation; confirm the actual billing and model-specific terms before committing.
PRICING & RIGHTS
Web tokens and API balance are two different budgets.
Pricing and terms last checked: Sep 7, 2026
- Free provides 150 daily tokens, not 150 images. Flow State has a separate free-use limit whose exact allowance was not confirmed.
- The official yearly-billing explanation lists Essential at $10/month with 8,500 monthly fast tokens: $120 for 12 months before applicable tax, calculated rather than checked at checkout. This is not a $10 month-to-month offer.
- Paid tiers add allowances and features such as private generation. Top-ups require an active subscription; rollover is capped. Relaxed generation covers selected models, not Flow State.
- Unverified month-to-month quotes are omitted. API PAYG is separately funded, and its calculator requires login; we have no verified price for a specified API image task.
COST TO OUTPUT
What an exploration budget can produce
Illustration only: Essential's officially stated yearly rate ($10/month equivalent, 8,500 monthly tokens) and Flow State at 1 token per new image. A workflow is 8 initial candidates + 8 similar variants, with existing results saved/downloaded. No other edits or upscaling are included.
O05O06O07O08O20O23$10 ÷ 8,500 × 1. This is an allowance allocation, not a cash price at which one image can be purchased.
$10 ÷ 8,500 × (8 + 8) ≈ $0.018824. New images consume tokens whether you keep them or not; saving existing results does not add this charge.
floor(8,500 ÷ 16), if every monthly token goes to this exact workflow. These are candidate sets, not 531 accepted assets. Annual billing does not imply all annual tokens arrive at once.
Standard generation, a chosen edit and an upscale need their own quotes; the 1-token rate cannot be reused. Accepted-asset cost also includes discarded attempts and human work. API pricing is separate.
The example requires an annual commitment of $120 before applicable tax, not a $0.019 purchase. Amounts retain the site's $ notation; checkout currency, regional charges and payment were not verified. Unused allowance, extra features, retries and cleanup increase effective cost. This is not a recommendation to buy annually or a promise of usable assets.
Commercial-use condition
Official help permits commercial use of free Web output, but the Terms assign its output rights to Leonardo. Paid output rights vest in the user only to the extent the law permits. Public content carries broad reuse permissions; paid private content has narrower service-processing terms. Check Private Mode and input rights. API trial/PAYG-only status and third-party/Beta conditions need separate confirmation; commercial permission is not a copyright or non-infringement guarantee.
Check the applicable plan and contract before purchase or commercial release. Web subscription, API usage and the external agent's own model costs are separate; do not infer a refund for every discarded or failed result.
PRODUCTION RISKS
Resolve these before production use.
Privacy is not the same as offline processing
Check the actual public/private setting before uploading. The private-content training restriction is not a zero-retention or local-processing promise; cloud providers, records and contractual exceptions still matter.
O10O14Budget for changes, not only selected downloads
A paid user reported unexpected submissions and changed settings. That is a dated account, not proof of a current universal bug. Recheck model, size and quote each time; top-up consumption-order conflicts remain unresolved.
O08O11U06Keep the API key outside the game
Use a controlled local tool or server for paid requests. Save the generation ID and inspect status: an empty result while pending is not a reason to repeatedly submit another paid generation. Clients need error handling and spending controls.
O16O17O21Exported images still need art direction
Inspect edges, text, visual scale and repeated character details in context. A personal trained model cannot be downloaded, and neither a successful image nor an old asset-pack case proves an entire current animation pipeline.
O04U01U02U03NOT VERIFIED
Claims this page does not make
- Interactive monthly billing state, checkout currency/tax, top-up pack prices and conflicting consumption order.
- Model-specific API quotes, trial/PAYG-only contractual rights and actual failed-task charges.
- Current cross-frame consistency, accepted-asset cost, throughput and Codex/Claude Code MCP client reliability.
RESEARCH METHOD
Research-reviewed from public evidence
Six concrete usage accounts across GameDev.net, Polycount, Reddit and iXBT; Reddit supplies three of six. Forum records and an independent walkthrough cover project use, iteration and friction. A historical asset pack and an undated mixed-tool animation failure retain those limits. Vendor tutorials, duplicated posts and unreadable review summaries were excluded. No generation, model training, engine import or authenticated API/MCP call was performed.
Research windowReviewed September 7, 2026. Most dated usage records are from March–August 2026, alongside a 2023 asset project and an undated animation attempt. They do not establish current model success rates.
OFFICIAL SOURCES20
- O01Leonardo.AiChecked Sep 7, 2026
Product scope; vendor quality and productivity claims are not independent evidence.
- O02How to Generate Images with Leonardo.AiChecked Sep 7, 2026
Browser generation, settings, the displayed token quote and downloads.
- O04Leonardo.Ai General FAQChecked Sep 7, 2026
Basic access, model portability, content rights and separate Web/API plans; older payment details are not a current guarantee.
- O05Flow StateChecked Sep 7, 2026
One token per newly generated image; saving existing results is free. Free-use limits apply; contradictory toggle directions are not repeated here.
- O06Leonardo.Ai PricingChecked Sep 7, 2026
Current tier names, allowances and features. Billing-toggle state was not interactively verified; the page omits unverified monthly quotes.
- O07Leonardo pricing: explicitly yearly-billed figuresChecked Sep 7, 2026
Only the clearly labelled Leonardo yearly pricing is used. Competitor comparisons and claimed winners are excluded.
- O08Tokens — Frequently Asked QuestionsChecked Sep 7, 2026
Task-dependent metering and restricted relaxed generation; one token is not a universal image price.
- O09Token RolloverChecked Sep 7, 2026
Rollover bank caps and the effects of plan changes; unused allowance is not unlimited savings.
- O10Leonardo.Ai Terms of ServiceChecked Sep 7, 2026
Terms dated January 19, 2026: free/paid output rights, public/private processing, subscriptions and Beta conditions.
- O11Top-up TokensChecked Sep 7, 2026
Active-subscription requirement; conflicting consumption-order text remains unresolved.
- O12Leonardo.Ai APIChecked Sep 7, 2026
Self-serve API and vendor commercial/privacy statements. Trial and PAYG-only rights are not assumed identical to paid private Web rights.
- O13Commercial UsageChecked Sep 7, 2026
Commercial permission for free and paid outputs does not mean identical ownership or exclusivity.
- O14Leonardo.Ai Privacy PolicyChecked Sep 7, 2026
Policy dated January 19, 2026; service processing, sharing and retention must be distinguished from training controls.
- O16API Quick Start GuideChecked Sep 7, 2026
Production API key, Bearer authentication, separate billing and generation ID after submission.
- O17API General FAQsChecked Sep 7, 2026
Pending/completed/failed states and result URLs; an empty image array may mean the task is still running.
- O18Official Python SDK: ImageChecked Sep 7, 2026
Documented generation and lookup methods; inspected as documentation, not installed or executed.
- O20API PAYG GuideChecked Sep 7, 2026
API balance and self-serve top-ups are separate from the Web token subscription.
- O21API Pricing and Plans FAQChecked Sep 7, 2026
Current PAYG versus legacy plans and optional automatic top-ups; example amounts are not task quotes.
- O23Plan with the Pricing CalculatorChecked Sep 7, 2026
The linked calculator required login; no authenticated model-specific task quote was obtained.
- O25Official Leonardo.Ai MCP ServerChecked Sep 7, 2026
Remote MCP, API-Key header and one documented generation tool covering three models; client compatibility was not tested.
INDEPENDENT EVIDENCE RECORDS6
- U01GameDev.net: Generating walking/running/jumping spritesChecked Sep 7, 2026
Year unknown. A multi-frame attempt using Copilot and Leonardo did not satisfy the author; separate tool results are unclear. The sketch is hand-drawn, not AI output. Vendor relationship unknown.
- U02Polycount: Newsstand prop packChecked Sep 7, 2026
August 2023 historical asset-pack project using reference-led magazine art. The author promotes the pack; other tools also contributed. Not a current-model benchmark; vendor relationship unknown.
- U03Reddit: Altmon made with Leonardo.AiChecked Sep 7, 2026
August 2026 character iterations from original artwork with Nano Banana. The author promotes a game in development; v4 is not proof of only four attempts. Release use and vendor relationship unconfirmed.
- U04Reddit: Chronicles of the Dying development accountChecked Sep 7, 2026
March 2026 self-reported mixed-tool game project. The author clarifies static images and short video, not Leonardo 3D models. This page uses the static-art account; the game was not run. Self-promotion; vendor relationship unknown.
- U05iXBT: Leonardo.Ai workflow walkthroughChecked Sep 7, 2026
Paid-account walkthrough of prompts, visual errors, Flow State and selection. Old plan names and subjective model rankings are excluded. Media commercial relationship unknown.
- U06Reddit: Feedback for Leonardo.ai DevsChecked Sep 7, 2026
March 2026 paid blog-illustration user reports accidental submissions and setting changes. Not a game task or proof those issues persist today; vendor relationship unknown.
DISCLOSURE
How this research was supported
No affiliate link, sponsorship, supplied account or gifted credits were used for this review. Some sources promote their own projects, and their vendor relationships are unknown. The research and both scoped selection judgments passed Owner Review on September 7, 2026.
This public research review will be revisited when pricing, terms, product versions, or material new evidence changes.