Which model should I use?
The Studio offers several models, and they are genuinely different — not better and worse versions of one thing. If a picture came out wrong, the next model along will often simply do it.
Start free, find the wording, then spend a credit on the version you want to keep.
Igarus
Free · unlimitedOur tuned default · powered by FLUX
Everyday portraits and scenes while you are still finding the prompt. It is free and unlimited, so it is the one to iterate on.
Least detail of the five, and it can lose small things — hands, insignia, lettering on a banner. Find the wording here, then spend a credit to finish it.
Made with Igarus
FLUX.2 pro
1 creditFast, crisp, great detail
Illustration. Crisp linework and strong colour, the closest to a drawn sourcebook plate — a good default for NPCs and monsters.
Leans stylised. If you wanted a photograph of a person, this will hand you a painting of one.
Made with FLUX.2 pro
Grok Imagine
1 creditxAI · photoreal & cinematic
Photoreal and cinematic. Faces, lighting and atmosphere — the one to reach for when a character should look like someone you could meet.
The same realism works against you for anything deliberately fantastical; a dragon comes out looking like a film still rather than an illustration.
Made with Grok Imagine
Nano Banana 2
2 creditsGoogle · natural & consistent
Consistency, and the only model that accepts a reference image. Attach a portrait you already have and it will keep the same character across new scenes.
Two credits, and it is the most literal of the set — a vague prompt gets a plain result rather than an interesting guess.
Accepts a reference image — attach a portrait and it keeps the same character.
Made with Nano Banana 2
GPT Image 1.5
4 creditsHigh quality · great value
Complicated prompts. It follows long instructions closely and is the most reliable at readable text — maps, signage, book covers, a labelled item.
Four credits, and slower than the rungs below it. Worth it for the piece you intend to keep, not for a first attempt.
Made with GPT Image 1.5
GPT Image 2
6 creditsBest prompt-adherence & text
The strongest prompt-adherence and text rendering available, for the piece that has to be exactly right.
Six credits, and currently not offered in the picker: it is too slow to generate while you wait. It returns once background generation exists.
Not currently in the picker. Listed so older art still says what made it.
Every image FTTHelper makes says which model made it, on the art itself. Open the Studio to try one.