Two months building a game asset tool because I couldn't draw the same character twice
I wanted to make a 2D action RPG.
I never got to the game part. I got stuck on assets and stayed stuck.
Buying asset packs didn't work — the art style never matched between packs, and a character from one pack next to a UI from another looked like two different games stitched together. Drawing them myself was worse. I'd finish a character, then need the same character from eight angles, and by the fourth angle the face had drifted into someone else. Then motion sheets. Then effects. Then UI. The game itself kept moving further away.
So I stopped making the game and made a tool instead.
You upload one character image. It generates eight directions in the same face, motion sheets from those, and effects, UI, backgrounds, tiles and sound in a matching style. You pick which frames you want and export as PNG sheets or JSON.
What was actually hard
Not the idea. The parts I didn't know anything about.
The image generation side was one of them. Getting eight angles of the same character is not the same problem as generating eight characters — consistency is the whole product, and I spent a long time on the gap between "this looks fine" and "this is the same character."
The server side was the other. I'm an artist, not a backend developer. Billing was the part that humbled me most. Credits, expiry dates, what happens when someone upgrades mid-cycle, what happens when you refund. Every one of those was a decision with no obviously right answer, and most of them I got wrong at least once before getting them right.
I also shipped a bug where the checkout button silently did nothing. The error message was rendering inside a hidden element, so the failure was invisible. It had been broken for days. I only found it because I went looking for something else.
Where I am now
Two months of building. Launched about a week ago.
5 users. 0 paying.
That's the honest number. I've posted on a few Discord servers, itch.io and a Korean indie dev forum this week. The Korean forum brought about 20 people in a day. The English-language channels brought exactly zero — my Discord posts got reactions and no clicks, and my X account has no followers so nothing moves.
What I'd do differently
Start talking about it while building, not after. I have no audience because I built in silence for two months.
Someone here told me this week to lead with the offer instead of the ask, and that landed hard. Everything I've posted has been an ask.
If you've been through the same wall — solo, technical product, zero distribution, and your only working channel is in the wrong language for your market — I'd like to hear what moved the needle for you.
The tool is at kitforge2d.com if you want to poke at it. Free credits on signup.
The Korean forum vs. English channels is a useful split. Are the people who actually click through behaving differently from the people who only react, or is the evidence still too thin?
The consistency problem sounds like the real challenge here. Generating assets is one thing, but keeping the same character recognizable across angles and animations seems much harder. Also, 5 users with 0 paying after a week is early, so I’d focus on figuring out which users actually get enough value to pay before changing too much.
Your buyer is probably not the person you were when you started. A hobbyist with an unlimited timeline solves the consistency problem by drawing slower, but a small studio or a jam team with a ship date cannot, and that deadline is what turns this from a nice tool into a purchase. Since consistency is the entire product, make the proof the marketing: one character in eight angles side by side, posted where teams with deadlines already are, will sell this better than any description of the tool.
The consistency check is the product, not the model. Freeze one reference sheet (front, three-quarter, side, two expressions) and reject a new direction if any of three game-facing things fail: silhouette, a small locked palette, or the origin/pivot. Frames that look fine but drift on those still break animation.
Ship the export contract before more generators. Lock canvas size, frame count, and origin in the PNG sheet or JSON so a Godot or Unity importer can drop it in without a trim pass. If that contract is loose, the artist still does the work the tool was supposed to remove.
On the English-channel silence: one failed-versus-locked clip in an existing engine thread (Godot, RPG Maker) beats another general launch post. People reply to a broken sprite, not to an announcement.
The 5 people who burned credits and never came back is not a traffic problem yet. They never saw the product you spent two months on. "Upload a character, get pictures" is how they score you against a general image model and leave. The actual product is "same face, eight directions, and the rest of the kit in that style." Put that comparison in the first session, not the changelog.
One 20-second clip of a face drifting versus locked, posted where people already fight that (Godot and RPG Maker servers, itch threads looking for an artist), will move more than a general launch post. Don't rebuild the site for language first. Caption that one clip in the language your buyers pay in, even if the channel you already have isn't that language. For a consistency tool, the side-by-side is the distribution. The build diary can wait.
The “I built in silence for two months” part is probably the most relatable thing here.
It’s interesting how building the product and building distribution can feel like completely different jobs — especially when you’re still figuring out who actually needs it.
I’m curious about one thing though: now that you’ve launched, are you changing the product based on what those first 5 users actually do, or mainly focusing on finding more people first?
Honestly, neither — because there isn't much to look at yet.
The 5 users signed up, spent their free credits on a few generations,
and that was it. Nobody came back. So there's no usage pattern to
learn from, just a handful of first tries.
The frame picker is quietly the most valuable thing you built, because it already measures the problem you said was hardest. Every time somebody keeps three of eight angles and discards the rest, that is a consistency score you did not have to define, collected on real characters instead of your own test cases. Kept frames per generation would tell you whether the model improved after a change and which styles it fails on, and it is also the honest unit to charge on, since nobody wants to pay for the five that drifted. Given that billing was the part that humbled you, pricing on what people actually keep sidesteps a lot of the refund questions too.
Thanks for taking the time to write this out properly — this is the
most useful thing anyone's said to me about the product.
The measurement half is going straight onto my list. Kept frames per
generation is sitting right there and I'm not recording it. I've been
judging model changes by eyeballing output, which is exactly as
reliable as it sounds. I'd have missed that entirely.
The pricing half I'm less sure about, and I'd like to poke at it with
you. Right now credits are spent at generation time, so charging on
kept frames means the cost is already sunk before the user decides.
I'd be eating the drift, which is fine as an incentive to make the
model better, but I can't tell yet whether that's a healthy pressure or
just a hole. And the obvious hole: what stops someone exporting the
sheet and then unpicking the frames?
Maybe the honest version is charging at generation like now, but
refunding credits for frames nobody kept. Same incentive, no gap to
walk through. Did you have something closer to that in mind, or were
you thinking genuinely pay-on-keep?