Overview

Leonardo AI is the image generation platform that game studios, indie developers, and concept artists quietly built their asset pipelines on top of. Founded in Sydney in late 2022 by JJ Fiasson, Chris Gillis, Jachin Bhasme, and Peter Runham — a team from the gaming and tech space rather than the ML-research pipeline — Leonardo launched publicly in early 2023 with a very specific pitch: not a general-purpose image toy, but a production tool for people who need to generate game assets, characters, environments, textures, and consistent illustrated style at scale. In mid-2026, that positioning has held. Canva acquired Leonardo in July 2024 for a reported $200+ million, and rather than absorbing the product, kept it running as an independent brand aimed at the pro creator segment Canva itself does not reach.

What sets Leonardo apart is a stack of workflow tooling built around the models rather than around a single prompt box. The core generator is one surface. Beyond it: fine-tune a custom model on your own art (native LoRA training with a real UI), lock a character across scenes, generate consistent-style asset runs, upload sketches for image-to-image with pose control, upscale to print resolution, and run all of it through a "Prompt Generation" helper that turns a one-line description into a working prompt. Leonardo is the AI image tool that acts like a design app, not a research demo.

The model lineup in 2026 covers Leonardo's own trained models (Phoenix, Anime XL, Vision XL, Diffusion XL, Kino XL) plus integrated access to Flux, and community-uploaded fine-tunes browsable inside the app. Phoenix, launched in late 2024 and iterated through 2026, is Leonardo's flagship own-model — strong prompt fidelity, solid photorealism, and pipeline-friendly seed control that matters more than benchmark headlines when you are shipping 200 assets for a game.

Leonardo the product has always over-indexed on gaming and illustration workflows, and the current subscription tiers reflect that — the sweet spot is a solo game dev, a small studio, or a concept artist producing dozens to hundreds of assets per week where consistency and asset-run quality matter more than one hero image.

Key Features

Flexible model roster. Phoenix as the flagship own-model, plus Anime XL for stylized illustration, Kino XL for cinematic, Vision XL for photographic, Lightning XL for fast iteration, and integrated access to Flux Dev and Flux Schnell. Leonardo is one of the few tools in 2026 where you can generate on Flux and Phoenix side-by-side in the same UI on the same prompt.
Native LoRA fine-tuning. Upload 8-30 reference images of a character, a style, or a product; Leonardo trains a LoRA in the background; use it in prompts by dropping it in from your model library. No ComfyUI setup, no rented GPU, no Python. This is the feature that keeps game studios and illustration teams on the platform — brand or IP consistency without owning a training pipeline.
Character reference and style consistency. Leonardo supports style reference images comparable to Midjourney's --sref and character-reference workflows for holding a face or figure across scenes. Combined with LoRAs, this is the strongest consistency toolkit outside of self-hosted Stable Diffusion.
Canvas Editor. Full inpainting, outpainting, masking, and layer-aware editing in a real canvas rather than a chat-style prompt box. Sketch an area, drop a reference, mask a region, regenerate a chunk — the flow is closer to Photoshop than to Midjourney's Vary Region.
Universal Upscaler. Multi-model upscaler with modes tuned for illustration, photography, and CG. Push a 1024x1024 output to 4K or 8K cleanly enough for print or in-game use.
Prompt Generation and Prompt Enhancement. Type a one-line intent, get a full prompt back with style, lighting, and composition detail. Useful for onboarding, occasionally useful for experienced users as a variation-generator.
Image-to-image with pose and depth control. Upload a sketch or photo, set a strength value, generate variations that respect the pose, composition, or depth of the input. ControlNet-flavored pipelines exposed through a real UI.
API for developers. cloud.leonardo.ai/api offers self-serve API access to Phoenix, Flux, and the specialized models with token-based pricing. One of the few platforms where you can hit both Flux and a well-tuned own-model behind a single API key.

Pricing

Leonardo prices in monthly tokens plus tier-based feature gates. All paid tiers grant commercial rights.

Plan	Monthly	Annual (per month)	Tokens/month	Approx images	Custom LoRAs
Free	$0	$0	150/day	~30-60/day	0
Apprentice	$12	$10	8,500	~1,700	1
Artisan	$30	$24	25,000	~5,000	5
Maestro	$60	$48	60,000	~12,000	10

Tokens are consumed based on model, resolution, and image count. A standard 1024x1024 Phoenix generation is roughly 4-5 tokens; a Flux Dev generation is roughly 10-14 tokens; upscaling costs a separate small charge. The "approx images" column above assumes a mix of default settings.

The free tier is generous by industry standards — 150 tokens per day, enough for real evaluation of the models on your own workflow before subscribing. Apprentice at $12/month is a reasonable entry for hobbyist game devs and casual illustrators. Artisan at $30/month matches Midjourney Standard on price and is the sweet spot for solo pros and small studios — 25,000 tokens plus five custom LoRAs covers a real production pipeline. Maestro at $60 is for high-volume asset producers or two-to-three-person studios.

API pricing is separate, token-based, and roughly comparable to Replicate for the same underlying models.

Pros and Cons

Pros

Native LoRA training with a working UI is genuinely rare and genuinely useful for anyone who needs consistent characters, styles, or IP assets.
Model roster is the deepest in the industry — Phoenix, Flux, Anime XL, Kino XL, community fine-tunes, all in one UI at one price.
Canvas Editor is closer to a design tool than a prompt box; art directors and illustrators feel at home in it.
Free tier is real enough for meaningful evaluation.
API access is self-serve and includes both Leonardo's own models and Flux.
Pricing is transparent and token-based rather than tied to opaque "GPU hour" metering.
Commercial rights on all paid tiers.

Cons

Aesthetic ceiling on any single model trails the specialists — Midjourney is prettier out-of-the-box, Flux is more photorealistic, Ideogram beats it on text. Leonardo wins on breadth and workflow, not on a peak-quality benchmark.
Interface is more complex than Midjourney or Ideogram — the depth of feature set costs onboarding time.
Text rendering is decent on Phoenix but still noticeably behind Ideogram.
Token accounting can feel opaque until you have run a few dozen jobs and internalized the cost model.
Canva acquisition has not yet meaningfully changed the product roadmap for good or for ill, but the strategic direction is a fair concern for pro users.

Best Use Cases

Indie game developers and small studios producing character portraits, environment concept art, item and prop assets, and consistent illustrated style at volume. LoRA fine-tuning plus Phoenix plus Canvas Editor is the pipeline this platform was built for.
Concept artists and illustrators doing style-consistent asset runs for clients — book covers, editorial illustration series, character design decks, world-building visuals.
Tabletop and TTRPG creators producing character portraits, monster illustrations, item icons, and map artwork.
Comic and manga creators using LoRAs to lock a character across pages and Canvas Editor for panel refinement.
Marketing and e-commerce teams who need brand-consistent asset runs at production volume and want a real editor UI rather than a prompt-only workflow.
Product builders wanting a single API that fronts Flux plus specialized models with commercial-friendly licensing.

Alternatives

Midjourney. Higher aesthetic ceiling for hero images and editorial work; better --sref for one-off consistency. No LoRA training, no canvas editor, no API. Use Midjourney when peak beauty matters and Leonardo when workflow and consistency matter. $10-120/month.
Flux. Available inside Leonardo as one of the model choices, or direct via api.bfl.ai and Replicate. Higher realism ceiling than Phoenix. If you only need one model and it needs to be photographic, Flux direct is cheaper. If you need multi-model workflow, LoRAs, and an editor, Leonardo wraps Flux better than the raw API.
Stable Diffusion via ComfyUI. The self-hosted alternative for anyone technical enough to set it up. Deeper control surface (ControlNet, IP-Adapter, thousands of community LoRAs), but the setup cost is real and there is no managed UI. Free to self-host; API from ~$0.002/image.
Ideogram. Beats Leonardo on text-in-image. If your workflow is typography-heavy, run Ideogram alongside Leonardo. $8-60/month.

For a deeper comparison across the three big image models, see our full write-up: Midjourney vs DALL-E 3 vs Stable Diffusion.

Getting Started

Sign up at leonardo.ai. The free tier's 150 daily tokens are enough to generate 30-60 images on Phoenix or a similar amount on Flux Dev — real evaluation before you subscribe.
Generate your first image on Phoenix. Type a prompt, pick Phoenix as the model, hit generate. Compare the same prompt on Flux Dev to feel the difference between the two engines.
Train your first LoRA. Upload 15-20 reference images of a character, style, or product, name the LoRA, and start training. The job runs in the background and completes in 20-40 minutes. Once trained, you use it by selecting it in your prompt panel.
Explore the Canvas Editor. Open an existing generation in Canvas, mask an area, and inpaint a change. This is where Leonardo diverges most from Midjourney and DALL-E.

FAQ

Is Leonardo better than Midjourney? For workflow, LoRA training, and asset consistency — yes. For raw aesthetic ceiling on a hero image — no, Midjourney still wins. Working pros run both.

Can I use Leonardo output commercially? Yes, on all paid tiers. The free tier is licensed for personal and non-commercial use only.

How much does LoRA training cost? LoRA training is included in the paid tiers (1 slot on Apprentice, 5 on Artisan, 10 on Maestro). Training a LoRA consumes a fixed batch of tokens per job.

Does Leonardo have an API? Yes, self-serve at cloud.leonardo.ai/api. It fronts both Leonardo's own models (Phoenix, Anime XL, Kino XL) and Flux.

Is Leonardo good for photorealism? It is competent, especially on Phoenix and via Flux integration. For pure photorealism at the highest quality, Flux direct is still marginally sharper. For photorealism inside a broader workflow, Leonardo is more than sufficient.

Did Canva change the product after the acquisition? As of mid-2026, no. Leonardo runs as an independent brand and the pro-creator positioning has held. The strategic direction is worth watching but has not degraded the product.

Verdict

Leonardo at $30/month on the Artisan tier is the correct default for anyone producing image assets at production volume rather than one-off hero images — indie game devs, illustrators, concept artists, comic creators, and small studios in particular. Native LoRA training plus the Canvas Editor plus a deep model roster including Flux is a stack that no other single tool matches at this price. Solo pros doing occasional generation are better served by Midjourney or Ideogram. Technical teams with ComfyUI expertise will prefer self-hosted Stable Diffusion or Flux for total control. Everyone in the middle — professional creators who need workflow, consistency, and a real editor without wanting to become an ML engineer — should look at Leonardo first. Canva's acquisition has not degraded the product; if anything, it has stabilized the roadmap. Skip Apprentice unless you are hobby-only; Maestro is only necessary at real studio volume.