Agent or manual generation
Start in conversation and let Layer select a route, or choose the model, references, dimensions and batch size yourself in the generation form.
O02O03O04Layer
Generate, organize and automate game-art candidates in one cloud workspace.
Evidence statusPublic-evidence reviewMedium confidenceLast checked Sep 22, 2026By MakeGameWithAI
PRODUCT OVERVIEW
What Layer is—and how it works.
Layer is a cloud creative workspace for game and entertainment teams. In its current Agent-first app, you can describe an asset in plain language, let the assistant choose a model and settings, compare results on a visual board, refine them and keep the useful files in a shared library. A separate generation form exposes model, reference, size and batch controls when you want to make those decisions yourself.
The product is broader than a single image generator. Projects group related sessions and assets; Reference Sets collect reusable character, object, environment or style material; optional LoRA training can turn a curated set into a custom model; Workflows chain multiple generation or processing steps. Reference-guided generation and custom training are different paths—a first candidate does not require training.
Layer currently exposes image, video, 3D and audio models, plus playable-ad and marketing workflows. This review makes a narrower decision about pre-generated 2D concepts, props, icons and related variants. PNG, JPEG and single-layer PSD export are delivery paths, not proof that transparency, pixel alignment, animation consistency or game-engine import already meet a project's standard.
The same workspace can be reached programmatically. Layer's official remote MCP documents OAuth setup for Codex, Claude Code, Cursor and other clients; the REST API uses a Personal Access Token. Both can estimate a task, submit it, poll status and retrieve results while drawing from the workspace's Creative Units.
HOW YOU USE IT
Create a Layer account and workspace, then obtain or buy enough Creative Units. Free describes an account without an active subscription; it does not establish an unlimited free generation allowance. MCP uses workspace OAuth, while REST tokens inherit account permissions and belong only in a trusted environment.
Reusable project context
Projects, the asset library and Reference Sets keep related outputs and approved visual material together. They reduce repeated setup; they do not guarantee character or style consistency.
O06O08Optional training and workflows
Curated references can become LoRA training data, while node-based workflows can repeat multi-step processing. Both add preparation, review and CU costs beyond a first image.
O07O17Official external-agent path
Remote MCP and REST cover discovery, estimates, asynchronous execution, status and results. The caller still needs authorization, a budget and human acceptance criteria.
O16O17O18O19O20INTERFACE & EXAMPLES
See how it works
Layer's official example shows the Agent alongside a board of character, backdrop, and potion assets. It is a vendor example, not a typical output claim.
Official help image from September 2026; app interface may change.Source: Layer Help Center
REVIEW SCOPE
A solo developer or small team exploring non-confidential 2D game-art candidates and optionally making that task callable from an external development agent. Current video, 3D, audio, playable-ad quality and production-scale SLA are outside this conclusion.
- Beginner friendlinessNot yet assessed
- The current browser workflow avoids local model deployment, node setup and mandatory training, but today's independent evidence is too sparse to label first-use onboarding easy or difficult. Confirm the available Creative Units and payment path before starting.
- Agent integrationEasy to integrate
- Official remote MCP and REST interfaces cover model and workspace discovery, matching price estimates, paid execution, status checks and result retrieval. Calls consume workspace Creative Units, and human budget and asset acceptance remain required.
Scope, integration requirements and sourcesChecked
Beginner scope: A browser user's first non-confidential 2D prop or icon candidate and export, excluding custom training and production delivery.
Agent scope: An external development agent generates a bounded 2D asset batch, polls the returned run and retrieves files; not autonomous approval or an entire game-art pipeline.
Confidence: learning Low · Agent Medium
- Layer remote MCPOfficial
- Layer REST APIOfficial
- Access
- MCP uses the remote Streamable HTTP endpoint and workspace OAuth; the official page includes Codex, Claude Code, Cursor and other client instructions. REST uses a Personal Access Token with the user's permissions and must stay in a trusted environment.
- Billing
- MCP and the app draw from the same workspace CU balance and add no Layer seat fee. The exact model, resolution, options and batch size determine the quote; the calling agent has separate costs. Public rollover and top-up-expiry rules conflict.
- Getting results
- Discover the workspace and model, estimate with the exact intended parameters, execute once, retain the inference ID, poll the matching run and download the returned assets. Explicitly set batch size and require approval before larger budgets or training.
- Limitations
- Easy means a documented end-to-end external task path, not guaranteed client compatibility, output quality or unattended production. Workspace roles, model availability, CU balance, rate limits, data handling and human review still apply.
These are independent editorial judgments, not an overall score or a claim of hands-on agent integration testing.
- Generate your first asset · Official · Learning evidence
- Layer AI historical walkthrough · Independent · Learning evidence
- G2 Layer review dated June 12, 2024 · Independent · Learning evidence
- Trustpilot reviews for layer.ai · Independent · Learning evidence
- Layer MCP setup and billing · Official · Integration evidence
- Layer MCP tool reference · Official · Integration evidence
- Layer forge price estimate · Official · Integration evidence
- Getting started with the Layer REST API · Official · Integration evidence
KEY FINDINGS
What the evidence supports—and what it does not.
Editorial inferenceMedium confidence
The workspace is the product difference
Layer combines model access with a board, projects, an asset library, Reference Sets, optional training and repeatable workflows. That can matter when a project produces many related candidates; it adds less value to a one-off image request.
O02O06O07O08Editorial inferenceLow confidence
A browser route lowers setup, but beginner ease remains unassessed
The documented first-asset path does not require a local GPU, node graph or custom training. Two positive records describe older versions, while a current reviewer did not complete a first image. That is enough to describe the path, not enough to label today's onboarding easy or difficult.
O03O04U01U02U03User reportsLow confidence
Exploration and precise correction are different tasks
The historical walkthrough reached multiple chest candidates and transparent export, but repeated local edits still missed the requested effect. Current tools have changed, so this supports a review checkpoint—not a current failure-rate claim.
U01Official factHigh confidence
Official automation covers a paid task lifecycle
MCP and REST expose workspace/model discovery, cost estimates, execution, status and result retrieval. An estimate must use the same parameters as execution; explicit batch size and budget limits prevent an agent from silently multiplying spend.
O16O17O18O19O20Editorial inferenceMedium confidence
Nominal candidate cost is calculable; accepted-asset cost is not
Current tariffs support a reproducible CU budget, but the evidence provides no current acceptance rate, retry distribution or cleanup time. A successful paid output that you discard still consumes the allowance.
O09O10O11EDITORIAL VERDICT
Our take
Medium confidenceLayer is worth considering when the real need is not merely one generated image, but a shared place to compare models, reuse game references, organize candidates and let a coding agent request assets. The official MCP/API path is unusually complete for this category. The weaker part of the evidence is the human experience: current independent reports do not establish easy onboarding, reliable style consistency or production-ready output. Begin with one bounded, non-confidential asset task and judge accepted files—not the number of generations.
Conclusion scopeMedium-confidence public-evidence assessment of the current 2D candidate and external-agent workflow. It is not a quality ranking or a claim about current success rates.
BETTER FIT
Worth auditioning when
- Solo developers or small teams that want multiple creative models behind one workspace and budget.
- Projects that repeatedly reuse characters, props, environments or a visual direction across art and marketing candidates.
- Codex, Claude Code or other MCP users who want an official estimate → generate → retrieve asset loop.
- Teams willing to review, refine and archive accepted outputs instead of treating every successful generation as game-ready.
POORER FIT
Do not depend on it yet when
- Anyone who needs a proven zero-payment first generation or an onboarding experience already supported by broad current user evidence.
- Projects requiring fully offline processing, zero third-party model handling or a blanket zero-retention guarantee.
- A pipeline that expects every image, sprite frame, 3D model or PSD to enter an engine without art and technical review.
- Buyers who need settled rollover, top-up expiry and cancellation terms before checkout without obtaining clarification.
WORKFLOW FIT
Where it fits in a production workflow.
Create a small reference-led candidate set
Choose one prop or icon, write its intended game use, dimensions and visual constraints, then generate a bounded number of options with either the Agent or form.
GuardrailUse original or licensed, non-confidential references. A Reference Set can guide generation without immediately starting paid custom training.
Review, refine and export one accepted file
Compare the candidates together, select one, make only the edits the task needs and inspect the exported PNG, JPEG or PSD at its actual game size.
GuardrailTreat single-layer PSD, transparency and successful download as separate from correct layers, clean edges, pixel alignment and engine readiness.
Automate only after the manual brief is stable
Connect the official MCP or REST route, discover the exact model, estimate with explicit batch parameters, execute once, poll the returned ID and save accepted results with their cost metadata.
GuardrailRequire approval for training, larger batches or a higher task budget. Never expose a PAT in game clients or public repositories.
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 game prop before building a style pipeline
This is a low-risk starting workflow synthesized from the evidence.
- 01
Pick one non-confidential prop or icon and define its target size, background, palette and acceptance checks.
- 02
Create an account, inspect the available CU balance and checkout conditions, and set a cash/CU limit before generating.
- 03
Use an existing model first. Generate a small explicit batch; do not treat custom training as an entry requirement.
- 04
Select, refine and export one candidate. Check edges, scale, readability and file behavior in the real game context.
- 05
Record payment, CU consumed, discarded outputs and cleanup time. Connect MCP/API only when the brief and approval rules are stable.
PRICING & RIGHTS
What $10 and 300 Creative Units mean for 2D candidates
Pricing and terms last checked: Sep 22, 2026
- The public entry subscription remains $10 per month for 300 CU. Layer has no per-seat charge and pools usage by workspace. A Free account has core access and can obtain one-off or reward units, but no stable automatic signup allowance was confirmed; $0 at the calculator's zero-usage state is not unlimited free generation.
- Seedream 4.0 is listed at 1.2 CU per generation, and Ideogram Remove Background at 0.3 CU. Model, resolution, duration and settings can change the tariff; use the matching estimate operation immediately before an automated paid call.
- The help center lists 500 one-off CU for $40 and 1,000 recurring additional CU for $60. These are separate pools with different nominal unit costs, not evidence that every subscription blends to the same rate.
- Successful generated outputs consume CU even when you reject them; a technical generation failure is documented as uncharged. MCP draws from the same workspace balance and does not add a seat fee, while the external coding agent has its own cost.
- Official documents conflict on monthly rollover and one-off-unit expiry. The examples below use only one current billing period and do not assume rollover, tax treatment, refunds or cross-period savings.
COST TO OUTPUT
From one candidate to a four-option prop workflow
Entry subscription allocation: $10 ÷ 300 CU = $0.033333… per CU. Fix the example to Seedream 4.0 at 1.2 CU per candidate and one separate 0.3-CU background-removal pass. These are nominal allocations inside the subscription, not separately purchasable finished-asset prices.
O09O10O11O18One successful model output before selection, correction or game-context checks. The cash entry remains the $10 subscription; Layer does not sell this candidate separately for four cents.
4 × 1.2 + 0.3 = 5.1 CU. Four options are a planned choice budget, not an observed 25% acceptance rate. Extra edits, upscales or training consume more.
floor(300 ÷ 1.2) = 250 single candidates. Alternatively, floor(300 ÷ 5.1) = 58 four-option workflows with 4.2 CU left. Do not add the two capacities together, and neither number is an accepted-asset count.
If the subscription is bought only for this task, cash paid is still $10 before tax, even when the chosen workflow allocates only $0.17. Actual cost also includes discarded successful outputs, cleanup, other tools and labor, divided by distinct accepted assets.
The model tariff and plan entry were rechecked on September 22, 2026. Accepted count, retry distribution and cleanup time remain unknown; if nothing passes review, no finite cost per accepted asset exists. Confirm the live estimate and checkout terms before spending.
Commercial-use condition
Layer grants access for lawful personal and commercial purposes. Subject to payment, input rights, applicable law and third-party model terms, Layer assigns whatever rights it has in generated content to the customer. Outputs may be similar to other users' results; this is not a promise of exclusivity, copyright eligibility or non-infringement.
The Terms say one-time CU do not expire and describe 10% rollover, while help articles say top-ups expire at cycle end and contain both no-rollover and up-to-10% language. Cancellation wording also differs. Do not stockpile units or depend on a cancellation date until the checkout/order terms that apply to your workspace are clear.
PRODUCTION RISKS
Resolve these before production use.
Billing rules conflict across official pages
Rollover, top-up expiry and cancellation notice cannot be reduced to one reliable sentence from the public material. Keep a screenshot or written confirmation of the terms shown to the purchasing workspace.
O10O11O13Successful output is not the same as accepted game art
The current independent set is too small and too old to establish style consistency, precise editing reliability or production acceptance rates. Review transparency, scale, anatomy, text, repeated characters and the full set in context.
U01U02U03No foundation-model training does not mean no processing
The Terms restrict training on customer inputs except customer-specific workspace improvements, but Layer and directly involved third-party providers still receive licenses needed to process, store and provide the service. Confidential IP needs the applicable plan, controls and contract review.
O13O15An agent can multiply spend and files quickly
Forge defaults and model settings can create batches. Estimate with exact parameters, set a task ceiling, poll the returned run instead of resubmitting and require human approval before training or scale changes.
O16O18O19NOT VERIFIED
Claims this page does not make
- Current beginner completion rate, first-generation payment friction and the amount of immediately usable free CU remain unknown.
- Current first-pass acceptance, character/style consistency, precise-edit success, cleanup time and cost per accepted 2D game asset are not established.
- Sprite-frame alignment, layered working files, animation continuity, engine import, runtime performance and production-scale reliability are outside the supported conclusion.
- The public conflict over rollover, one-off CU expiry and cancellation notice remains unresolved for an individual checkout.
- No comparative quality ranking is made against Scenario, Leonardo.Ai, Recraft, ComfyUI or other art pipelines.
RESEARCH METHOD
Research-reviewed from public evidence
We refreshed 20 official product, help, pricing, legal and developer sources. Thirteen independent environments were searched in the underlying research; three concrete usage records were retained across an independent design blog, G2 and Trustpilot, plus one community technical repository kept separately. Reddit, GameDev.net, Unity Discussions, Epic forums, Polycount, Blender Artists, Product Hunt, Hacker News and ArtStation did not add a qualified current-use case. Two retained records concern older versions, while the current Trustpilot record is a failed onboarding attempt rather than an output review. This is sufficient for a bounded product, cost and interface page—not a representative satisfaction sample or a current quality consensus.
Research windowPublic evidence reviewed September 7 and refreshed September 22, 2026
OFFICIAL SOURCES20
- O01LayerChecked Sep 22, 2026
Current product scope and positioning. Vendor productivity and quality claims are not treated as independent evidence.
- O02What is Layer?Checked Sep 22, 2026
Plain-language Agent workflow, visual board and supported modalities.
- O03Generate your first assetChecked Sep 22, 2026
Documented first-image path after the workspace has Creative Units.
- O04Generate and edit assets using the generation formChecked Sep 22, 2026
Manual model, reference, size and batch controls as an alternative to Agent mode.
- O05Exporting assets on LayerChecked Sep 22, 2026
PNG, JPEG and single-layer PSD export, batch ZIP and metadata controls; export availability is not engine-readiness proof.
- O06Using reference setsChecked Sep 22, 2026
Reusable character, object and style references, distinct from custom-model training.
- O07How to train a custom model (LoRA)Checked Sep 22, 2026
Training-data preparation and optional custom-style workflow; not required for a first generation.
- O08A tour of your workspaceChecked Sep 22, 2026
Current projects, sessions, boards, asset library, reference sets, workflows and models organization.
- O09Layer pricingChecked Sep 22, 2026
Current subscription entry, model CU tariffs and no-per-seat positioning. Interactive calculator defaults are not a quote.
- O10Guide to Layer subscriptions and pricingChecked Sep 22, 2026
Studio entry, top-ups, workspace pooling and conflicting rollover wording retained as written.
- O11How Creative Units workChecked Sep 22, 2026
Successful-output charging, failed-run treatment, usage logs and rollover wording.
- O12Layer rewards programChecked Sep 22, 2026
Earned and claimed units are conditional rewards, not an automatic signup allowance.
- O13Layer Terms of ServiceChecked Sep 22, 2026
Payment, generated-content rights, customer inputs, third-party models, training and rollover terms.
- O14Commercial useChecked Sep 22, 2026
General commercial permission subject to the current Terms and input rights.
- O15Layer Privacy PolicyChecked Sep 22, 2026
Service data processing, disclosure, retention and user choices; not a zero-retention promise.
- O16Layer MCP serverChecked Sep 22, 2026
Remote MCP URL, OAuth setup for Codex and other clients, shared workspace CU and task examples.
- O17MCP tool referenceChecked Sep 22, 2026
Current model, workflow, forge, file, workspace and instruction tool groups.
- O18estimate_forge_priceChecked Sep 22, 2026
Side-effect-free estimate with parameter parity before a paid forge call.
- O19execute_forgeChecked Sep 22, 2026
Asynchronous image, video, 3D or audio submission and explicit batch controls.
- O20Getting started with the Layer APIChecked Sep 22, 2026
REST base URL, Personal Access Token authentication and end-to-end generation documentation.
INDEPENDENT EVIDENCE RECORDS4
- U01Akane Lee: trying Layer AIChecked Sep 22, 2026
Historical Asset Studio walkthrough: chest candidates and transparent export were approachable, while precise edits needed repeated attempts. It does not establish the current Agent-first experience; vendor relationship unknown.
- U02G2: Layer review dated June 12, 2024Checked Sep 22, 2026
Historical small-business creative and marketing use; useful only as an older workflow signal. Not a game-production case; relationship unknown.
- U03Trustpilot reviews for layer.aiChecked Sep 22, 2026
One unprompted reviewer reported confusing onboarding and a card request before completing a first image. Account balance and exact route are unknown; it is a failed attempt, not an output-quality test.
- T01Community Layer AI MCP serverChecked Sep 22, 2026
Community wrapper retained as historical technical evidence; incomplete features and older assumptions mean it is not used in place of Layer's official MCP.
DISCLOSURE
How this research was supported
Independent editorial research. MakeGameWithAI used no affiliate link, sponsorship, vendor-provided account, free credits, interview or technical support for this review. Third-party relationships are unknown unless noted. The original scoped research and Agent decision passed Owner Review on September 7, 2026; the Owner approved making the public page on September 22, 2026. Publication does not remove the disclosed evidence and billing gaps.
This public research review will be revisited when pricing, terms, product versions, or material new evidence changes.
Related news
All news
Layer announces Playables for turning game assets into interactive ads
The September 29 announcement brings game art, conversational revisions and HTML export into Layer's playable-ad workflow. Generation uses Creative Units; exports have no separate fee.
Read article