Handy for a matched set of game-small assets. I go elsewhere for one cinematic still.
Leonardo’s menus feel built for twenty item icons, not one cover painting. I have used it for a shared stroke width on icons, props, and rough turnarounds. Style models and training make “looks like one game” more reliable than luck in a general chat. It can do landscapes; I rarely pay it for that — Midjourney is less work. Tokens vanish on upscales and bursts. Type is still weak; UI words get added later. Versus local SD you skip the environment and lose absolute control. Versus DALL·E you gain asset-oriented models and a batch feel. I lock canvas and alpha needs first so I do not spend tokens on a transparency that is not there. Edges still get cleaned in Aseprite or Photoshop. It is an asset shop, not an art director.
People shipping game or app icon sets. One-off poster jobs do not need a special plan.
Matched batches
One model for icons and props beats three websites.
Production sizes
Squares, small frames, tiles — closer to a pipeline than a default widescreen.
No local babysitting
A browser is enough for a round.
Training is there
A few samples can lean toward “our game.”
Tokens go fast
Upscales and bursts empty the pack before you feel started.
Weak cinematic one-off
I still open Midjourney for a single film frame.
Edges need hands
Alpha and jaggies rarely drop straight into an engine.
Lock style and size; start line-ish; decide color after.
Add “side view” as a draft, not an animation frame.
Freeze lighting words so noon and neon do not mix.
For style tests. Not for a full set.
Usually after edge and size cleanup. Treat it as source, not import.
Pose lock and privacy: local. A batch of small art this afternoon: Leonardo.
Set real type yourself. Generated glyphs are texture.
Yes on tiny, similar sets. Use a house model first, then train.
The link opens the Leonardo AI site so you can check our judgment. The review is written by AI WebLink.