Guides
Every piece here is built on images we generated and numbers we measured on our own hardware, with the prompts printed so you can check the work. Failures included.
- Model fileQwen-Image-3.0 has no open weights, so we tested the ones that doWe queried both model hubs on 2026-08-13, then ran six prompts aimed at the dense-text work 3.0 is known for through the open model we host.
- Prompt setTwelve action figure prompts, and what the packaging line actually buysThe toyification trend run through twelve prompts on our own hardware, with an A/B on the blister-pack clause everyone pastes and a warning about what the model reaches for when you do not name a brand.
- ExperimentAI crayon drawings come out neat, and bad photos come out rightTwelve prompts across four kinds of badness, run on 2026-08-13. The crayon scribbles came back competent and the flash-blown snapshots came back convincing.
- RightsAI toy images: ask for a generic one, get somebody else's productEight prompts, no brand named in any of them, run on 2026-08-13. Four came back as a design a toy buyer names on sight.
- RightsCan you sell AI-generated art?Yes. The harder question is whether you own it, and the answer to that one is mostly no.
- RightsIs AI art allowed on Etsy? Yes, and they want you to say it was AIThe rule that spread across selling platforms in 2026 is not a ban. It is a checkbox, and missing it is what gets listings pulled.
- GuideHow to make an AI imageThree prompts, ten seconds each, and the three requests that came back wrong so you know what not to ask for.
- ExperimentHow to write AI image prompts: one added word beats a rewriteFour scenes, three versions each: bare, one word added, and a thirty-word rewrite. The one word does most of the work.
- Prompt setAn AI image style list: 36 styles, one prompt, one image eachThe same lighthouse rendered 36 ways, so you can see which style words the model honours and which it quietly ignores.
- ExperimentHow to tell if an image is AI generated: counting fingers stopped workingWe probed six of the classic giveaways one at a time. Five of them are dead, and the one that survives is the one nobody mentions.
- GuideWhat inpainting is in AI, and how img2img differsTwo words that get used interchangeably and mean different things. One rebuilds your whole picture; the other is supposed to leave most of it alone.
- Model fileHow AI image generation works, in the ten seconds after you press generateThe wait is not the model thinking. It is a queue, a text encoder, a handful of denoising steps and a file transfer, and we can show you how long each one takes.
- ExperimentWhat a negative prompt is, and why it does almost nothing hereWe asked for twenty things to be left out. Four came out. Five got more prominent than they were before we objected.
- Prompt setAI tattoo ideas that survive contact with a tattoo artistEleven designs on clean white, and the five words that decide whether you get a stencil or a photo of somebody else's arm.
- ExperimentHow to make AI images look real, starting with what looks fakeFour scenes shot twice: once the way people usually ask, once with the flaws written in. The second version is not better lit. It is worse lit, on purpose.
- ExperimentWhy does AI art have a yellow tint? Ours does not have oneTwenty-four images of deliberately white things, measured rather than eyeballed. The average came out slightly cool.
- ExperimentWhat an AI image generator with no restrictions actually blocksSwimwear, lingerie and a boudoir set went straight through. A marble statue in a museum did not, because of the word standing next to it.
- TestedWhat Grok's spicy mode allows, and what it now costs to reach itA paid tier, an age check, a phone app and a country that has not blocked it. We ran the same register of prompts through a free generator to see what the gate is protecting.
- GuideWe ran ten outfit swaps through a free AI editor. The clothes changed and the shoes did not.One generated model, ten instructions, one refusal. What breaks is never the garment you named. It is everything you did not.
- ExperimentWe shot the same person twelve times. Six of them were not the same person.Two descriptions, six scenes each. What holds a face together between shots is the words you repeat, and one of the five things we asked for never turned up at all.
- TestedWe spent $0.32 to find out what Nano Banana Pro does that a free generator cannotEight prompts through both, one attempt each. The paid model wins every picture that contains words, and the gap on photographs is narrower than the price difference suggests.
- Model fileZ-Image Turbo: what it does well, and where it breaksThe model behind every image on this site, run through six prompts on our own hardware, with the failures left in.
- Model fileQwen Image Edit: what it actually changes when you edit one thingThe model behind our image editor, run through six instructions on pictures we made ourselves, with everything it changed by accident left in.
- ExperimentWhy AI struggles with hands, and the one pose that still breaks itWe generated 36 hands in three poses and scored every digit. Two of the three poses came back clean.
Ad